1008-00005 · Lumos Analytics Inc · Discovery

deposition-summary-yun.docx DOCUMENT Litigation

7 / 35 ← → ✕

Contract analysis Extract to table

PRIVILEGED AND CONFIDENTIAL — ATTORNEY WORK PRODUCT

CALDERWOOD & HARKNESS LLP intellectual Property Litigation Group

INTERNAL MEMORANDUM

Date: June 27, 2023

To: Wendy Schultz, Lead Associate, intellectual Property, Calderwood & Harkness LLP

From: Omar Siddiqui, Working Associate, intellectual Property, Calderwood & Harkness LLP

Re: Deposition Summary — Derek Yun (June 22, 2023)

Client: Lumos Analytics Inc.

Matter: Lumos Analytics Inc. v. NovaSpark Technologies Inc. et al., Case No. 23-CV-004871, Superior Court of the State of California, County of Santa Clara

C&H Matter Number: 1008-00005

I. Introduction and Deposition Logistics

This memorandum summarizes the deposition testimony of Derek Yun, taken on June 22, 2023, at Calderwood & Harkness LLP's San Francisco office located at 555 California Street, 40th Floor, San Francisco, California 94104. Mr. Yun is a former Machine Learning Engineer II at Lumos Analytics Inc. ("Lumos") and is currently employed as Founding Engineer at NovaSpark Technologies Inc. ("NovaSpark"). He is one of three individual defendants in this action. At the time of the deposition, Mr. Yun was 31 years of age.

The deponent was represented by Patricia Whitfield of Whitfield, Raines & Gerber LLP, located at 235 Montgomery Street, Suite 1800, San Francisco, California 94104. Samuel Gerber of the same firm was also present on behalf of the deponent throughout the proceeding.

The examination was conducted by Wendy Schultz on behalf of Lumos. I attended the deposition and took contemporaneous notes, which form the basis for this summary in conjunction with the rough transcript received from the court reporter on June 24, 2023. The final transcript runs approximately 287 pages and covers approximately 6.5 hours of testimony, inclusive of a morning break, a lunch break, and two additional short breaks requested by defense counsel.

Also present on behalf of Lumos was Naveen Kapoor, Senior Consultant at Meridian Cybersecurity Advisors Inc., who served as a technical advisor. Mr. Kapoor did not participate in the questioning but was available for consultation during breaks regarding technical issues arising from the testimony.

A certified court reporter was present and transcribed the proceedings. A videographer was also present; the deposition was recorded on video. The deposition was taken under oath pursuant to the California Code of Civil Procedure, and Mr. Yun was duly sworn before examination commenced.

This memorandum is organized thematically rather than in strict chronological order of the questioning, to facilitate review and use of the testimony in connection with upcoming motion practice, the mediation brief, and trial preparation.

II. Background and Employment at Lumos Analytics Inc.

A. Hire Date and Role

Mr. Yun confirmed that he joined Lumos Analytics Inc. in January 2021 as a Machine Learning Engineer II. He testified that his hiring process involved a multi-stage technical interview, including a take-home coding exercise, two live technical interviews, and a behavioral interview. He received an offer letter, which he accepted, and his first day of employment was January 11, 2021.

At Lumos, Yun worked within the ML Infrastructure team. He reported directly to a team lead, with his reporting chain extending ultimately to Rachel Fong, Vice President of Engineering. Yun testified that he worked closely on a day-to-day basis with Rajiv Anand, who held the title of Principal Engineer — ML Infrastructure and who served as the technical lead for the team. Yun described Anand as the most senior technical contributor on the ML Infrastructure team and indicated that he looked to Anand for guidance on architectural and design decisions.

When asked to describe his role and responsibilities, Yun stated that his work consisted primarily of developing and refining machine-learning models for the LumosEdge platform. Specifically, he described his core responsibilities as including feature engineering (the process of creating derived variables from raw data inputs to improve model performance), model training (the iterative process of running ML algorithms against training data), hyperparameter tuning (the optimization of model parameters that govern the training process itself), and model validation (the testing and evaluation of model outputs against held-out data to assess performance and generalizability). He testified that his day-to-day work involved extensive use of Python, TensorFlow, PyTorch, and Jupyter notebooks as primary development tools.

Yun acknowledged that the LumosEdge platform was Lumos's core commercial product and described it as an AI-driven analytics platform designed to generate financial trading signals from alternative data sources for use by hedge funds, asset managers, and proprietary trading desks.

B. Access to Proprietary Materials

Yun's testimony regarding his access to Lumos proprietary materials was detailed and is of particular significance to our trade-secret misappropriation claims. The following key admissions were obtained:

First, Yun acknowledged that he had direct access to Lumos's proprietary training datasets, which he confirmed comprised approximately 14.6 terabytes of historical financial and alternative data. He explained that these datasets included both traditional financial market data and alternative data streams drawn from range non-traditional sources, including satellite imagery analytics, web-scraping outputs, credit card transaction aggregations, and geolocation data. He confirmed that these datasets were compiled and curated by Lumos at substantial cost over several years and were not publicly available.

Second, Yun confirmed that he had access to model hyperparameter configurations for the LumosEdge ensemble ML model architecture. He described the hyperparameter configurations as representing "the result of extensive experimentation by the team, including myself" and acknowledged that these configurations were developed over many months of iterative work. He agreed that the hyperparameter configurations were closely guarded within Lumos and were not shared externally.

Third, Yun confirmed that he had access to Lumos's proprietary feature-engineering libraries, which he acknowledged contained 2,847 individual engineered features for financial signal generation. He described the feature-engineering libraries as "one of the most valuable parts of the LumosEdge platform" because the features represented a cumulative investment of engineering effort in identifying, constructing, and validating derived variables that served as inputs to the ML models. He acknowledged that many of these features were novel and were not derived from publicly available academic literature.

Fourth, Yun acknowledged familiarity with the DataStream ETL (Extract, Transform, Load) pipeline to the extent that it fed data into his ML workflows. He explained that DataStream was the data ingestion and processing system that prepared raw data from Lumos's various vendor integrations for consumption by the ML models. However, Yun stated that his access to the DataStream codebase itself was limited to read-only permissions and that the pipeline was primarily maintained by a separate engineering team. He testified that he interacted with DataStream primarily through its output — the cleaned, normalized, and structured data that arrived in the ML team's development environment.

Fifth, Yun testified that he had "some awareness" of the LumosEdge 3.0 product roadmap through attendance at team meetings and planning sessions. However, he claimed that he never personally possessed the full 37-page LumosEdge 3.0 roadmap document. He stated that roadmap discussions in team meetings tended to focus on ML-related features and timelines, and that he did not have visibility into business development, go-to-market, or pricing aspects of the roadmap. This testimony should be viewed in light of the fact that Rajiv Anand, who is alleged to have forwarded the full 37-page roadmap to a personal email account, was Yun's direct technical lead and primary collaborator.

C. Execution of the PIIAA

Yun confirmed that he executed Lumos's standard Proprietary Information and Inventions Assignment Agreement ("PIIAA") at the time of his hire in January 2021. He testified that the PIIAA was presented to him as part of his onboarding paperwork and that he signed it on or about his first day of employment.

When shown the executed copy of the PIIAA (marked as Exhibit 14 during the deposition), Yun identified his signature on the signature page and acknowledged the document as the agreement he had signed. He confirmed that the document appeared to be a complete and accurate copy of the agreement.

Under questioning, Yun acknowledged that the PIIAA contained the following key provisions:

(a) An assignment of all inventions clause, which assigned to Lumos all rights, title, and interest in any inventions, works of authorship, discoveries, developments, and improvements conceived or reduced to practice during his employment and within the scope of Lumos's business;

(b) A confidentiality and non-disclosure provision surviving termination of employment for a period of three years, pursuant to which Yun agreed not to disclose or use Lumos's proprietary and confidential information;

(c) A non-solicitation of employees clause for a period of 12 months following separation from employment, pursuant to which Yun agreed not to solicit or recruit Lumos employees; and

(d) A covenant not to use or disclose Lumos trade secrets at any time, which by its terms was perpetual and not limited to any post-employment period.

When questioned about his understanding of these restrictive covenant provisions, Yun stated that he "understood in general terms" what the PIIAA required of him. However, he claimed that he did not consult an attorney before signing the agreement and that he did not review the PIIAA again at any point during his employment, including before or after his resignation in October 2022. He characterized his review at the time of signing as "a quick read-through" and stated that he understood the agreement was "standard" and "everyone signed it."

Critically, Yun acknowledged that he did not return all company materials upon termination of his employment, as required by the PIIAA. He further acknowledged that he did not confirm the deletion of Lumos proprietary information from his personal devices or accounts at the time of his departure. The PIIAA contains an express provision requiring the return or certified destruction of all proprietary materials upon separation. Yun's failure to comply with this obligation is directly relevant to both our breach of contract and trade-secret misappropriation claims.

III. Circumstances of Resignation from Lumos

A. Decision to Leave

Yun testified that he first began considering leaving Lumos in approximately August or September 2022. He attributed his initial dissatisfaction to what he described as "limited career growth opportunities" and "a desire to be part of something more entrepreneurial." His testimony on this point was notably vague regarding specific timelines and triggering events.

More significantly, Yun acknowledged that Rajiv Anand approached him in approximately mid-September 2022 about joining a new startup venture, which ultimately became NovaSpark Technologies Inc. Yun testified that Anand described the opportunity in general terms as "building an AI-driven alternative-data analytics platform for hedge funds and trading desks." Yun stated that Anand's pitch appealed to him because of the opportunity to be a founding engineer and because of his respect for Anand's technical abilities.

Yun further testified that he also spoke with Serena Mallick about the NovaSpark opportunity before either of them resigned from Lumos. However, Yun was notably vague on the timing and substance of those conversations with Mallick. When pressed for dates and details, he responded with qualifications such as "I don't remember exactly when" and "we talked about it generally." He was unable or unwilling to specify whether Mallick approached him or he approached Mallick, or whether Anand served as an intermediary between them.

Throughout this portion of the examination, defense counsel Patricia Whitfield was particularly active. She objected to several questions regarding the content of conversations between Yun and Anand on attorney-client privilege grounds, specifically asserting that certain communications occurred after the parties had retained counsel in anticipation of the present litigation. These objections were noted on the record. At Ms. Whitfield's instruction, Yun declined to answer approximately three to four questions in this area. The privilege log implications of these instructions should be evaluated, as several of the withheld communications appear to predate the filing of this action and may not be properly subject to the attorney-client privilege.

B. Resignation

Yun resigned from Lumos on October 21, 2022, effective immediately. He did not provide two weeks' notice or any other advance notice period. He submitted his resignation via email to his direct manager and to Rachel Fong, Vice President of Engineering. When shown his resignation email (Exhibit 27), Yun confirmed its authenticity. The email was brief and provided no reason for his departure beyond a general statement that he had "decided to pursue a new opportunity."

Yun testified that he did not participate in a formal exit interview with Lumos's Human Resources department. He stated that he was not contacted by HR regarding an exit interview and that he did not initiate one. He acknowledged that the employee handbook, with which he was familiar, recommended that departing employees participate in an exit interview process.

Notably, Yun did not disclose at the time of his resignation that he intended to join NovaSpark or any other competing venture. He did not inform Lumos that his former colleague Rajiv Anand had recruited him, nor that he planned to begin work at a direct competitor within days of his departure.

Yun acknowledged that he was aware, at the time he submitted his own resignation, that Anand had already resigned from Lumos on September 30, 2022, and that Mallick had resigned on October 14, 2022. The coordinated timing of the three departures — September 30, October 14, and October 21, 2022, in order of seniority — is significant and will be addressed further in Section X below regarding the comparison with prior deposition testimony.

C. Commencement at NovaSpark

Yun confirmed that he began employment at NovaSpark Technologies Inc. on October 24, 2022, which was three calendar days after he resigned from Lumos and the next business day (October 22 and 23, 2022, fell on a Saturday and Sunday, respectively). His title at NovaSpark was Founding Engineer.

Yun confirmed that NovaSpark was incorporated in the State of Delaware on October 3, 2022, by Rajiv Anand. This incorporation date is notable: it falls only three days after Anand's resignation from Lumos on September 30, 2022, which strongly suggests that the incorporation was planned well in advance of Anand's departure. NovaSpark is headquartered at 88 Bluxome Street, Floor 4, San Francisco, California 94107.

When asked to describe NovaSpark's business, Yun acknowledged that NovaSpark's stated business — AI-driven alternative-data analytics for hedge funds and proprietary trading desks — overlaps directly with Lumos's core LumosEdge offering. Yun attempted to qualify this by stating that "the alternative data analytics space is large and there are many different approaches," but he could not identify any material differentiation between NovaSpark's stated business model and Lumos's when pressed on the point.

IV. The Jupyter Notebook Downloads — Key Exfiltration Testimony

This section addresses the most consequential portion of the deposition: Yun's testimony regarding the downloading of 47 proprietary Jupyter notebooks to his personal Larkspur Technologies account on October 18, 2022, three days before his resignation from Lumos. The testimony in this area followed a pattern of initial denial, incremental concession under confrontation with forensic evidence, and ultimately grudging acknowledgment of the core facts.

A. The October 18, 2022 Downloads

When shown Brindoval Digital Forensics LLC's Preliminary Forensic Report (Exhibit 22, dated May 1, 2023) and the corresponding Lumos IT access logs (Exhibit 23), Yun was questioned in detail about the download of 47 proprietary Jupyter notebooks to his personal Larkspur Technologies account on October 18, 2022.

Yun's testimony evolved significantly over the course of questioning on this topic. Initially, Yun testified that he "did not recall" downloading Jupyter notebooks to a personal Larkspur Technologies account. His exact words, as recorded in the rough transcript, were: "I don't recall doing that, no." When pressed on whether he had a personal Larkspur Technologies account, he acknowledged that he did but stated he "didn't use it for work."

When confronted with the forensic metadata contained in the Brindoval report — including specific timestamps corresponding to the downloads and his Lumos user credentials associated with the file-transfer events — Yun shifted his testimony. He stated that "it is possible" the downloads occurred but that he could not "specifically recall" initiating them.

After further questioning, Yun offered an alternative explanation. He acknowledged that he had a personal Larkspur Technologies account that was synced on his Lumos-issued laptop and stated that files "could have been transferred through that sync." He suggested that the Larkspur Technologies desktop client may have automatically synced files that were present in certain directories on his work machine without his affirmative action.

When this explanation was challenged — given that the Brindoval report documented the files as having been moved from specific Lumos internal directories to the Larkspur Technologies sync folder, rather than having been passively synced — Yun ultimately conceded that he "probably did" download the 47 Jupyter notebooks. However, he characterized the action as "inadvertent" and claimed that he was "backing up personal work notes" rather than deliberately taking proprietary materials. He stated: "I had a lot of notes mixed in with work stuff and I was just trying to save my personal things."

Yun was then shown a detailed list of the 47 notebook filenames (Exhibit 24). The filenames included clear and unmistakable references to Lumos-specific proprietary terms and projects, including but not limited to the following:

• Filenames referencing "LumosEdge" model configurations;

• Filenames referencing feature-engineering pipelines with Lumos-specific naming conventions;

• Filenames containing training data schema definitions tied to Lumos's proprietary data structures;

• Filenames referencing proprietary ensemble model architectures, including the "LE3" designation associated with the LumosEdge 3.0 development effort; and

• Filenames containing references to specific data vendor integration outputs processed through the DataStream pipeline.

When confronted with this list, Yun acknowledged that many of the filenames referenced Lumos-specific proprietary terms and project names. He conceded that it would be "obvious from the filenames" that the notebooks contained Lumos-related work. However, he maintained his position that some of the notebooks contained "his own personal notes and experiments" that he considered to be his personal intellectual property, distinct from Lumos's proprietary materials. He was unable to identify which specific notebooks he considered to be personal rather than proprietary.

B. Knowledge of Proprietary Nature

A substantial portion of the deposition was devoted to establishing Yun's knowledge and understanding that the 47 Jupyter notebooks contained proprietary information belonging to Lumos. His testimony on this point, while frequently qualified and hedged, yielded significant admissions.

Yun acknowledged that the Jupyter notebooks were created on Lumos systems, using Lumos computing resources (including Lumos's cloud-hosted development environment and GPU instances), during the course of his employment at Lumos and within the scope of his assigned duties. He agreed that, as a general matter, work product created using company resources during the course of employment belongs to the employer.

Yun further acknowledged that the notebooks contained feature-engineering code developed specifically for the LumosEdge platform. He conceded that the code in the notebooks referenced specific engineered features drawn from the 2,847-feature proprietary library maintained by Lumos's ML Infrastructure team. When asked whether those features were publicly available, Yun stated that "some of the underlying concepts are discussed in academic literature" but admitted that "the specific implementations and the specific features as engineered for LumosEdge were developed internally."

Yun admitted that the notebooks contained model hyperparameter configurations for the LumosEdge ensemble model. He described these configurations as representing "the result of extensive experimentation by the team, including myself" and acknowledged that they embodied significant computational effort and engineering judgment that was not replicable from public sources.

Yun further conceded that the notebooks referenced proprietary training data schemas — the structural definitions and metadata describing how Lumos organized and formatted its 14.6-terabyte training dataset — and that the notebooks contained references to data sourced from Lumos's 43 alternative-data vendor integrations processed through the DataStream ETL pipeline. He acknowledged that the identity of Lumos's data vendors, the specific data feeds obtained from those vendors, and the schemas used to process that data were confidential business information.

Despite these concessions, Yun repeatedly attempted to characterize portions of the work contained in the notebooks as "his own intellectual contribution" and suggested that some of the methods employed were based on "publicly available research papers." He referenced general concepts such as gradient boosting, attention mechanisms, and ensemble stacking as examples of publicly known ML techniques that he used in the notebooks. When counsel pointed out that the application and specific configuration of those techniques for financial signal generation from alternative data — the particular combination and optimization reflected in the notebooks — constituted proprietary Lumos work product, Yun did not directly dispute this characterization but restated his position that "there was personal work mixed in."

Critically, when asked directly whether any of the 47 downloaded Jupyter notebooks consisted entirely of publicly available material with no proprietary Lumos content, Yun could not identify a single one. He paused for an extended period before responding that he "would need to review each one individually" to make that determination, but he was unable to name even one notebook that he was confident contained no Lumos proprietary information.

C. Failure to Return or Delete Materials

Yun confirmed that he did not return the 47 Jupyter notebooks or any other Lumos materials upon his resignation on October 21, 2022. He acknowledged that the PIIAA he signed in January 2021 contained an express provision requiring the return of all company materials, including electronic files and copies thereof, upon termination of employment for any reason. He further confirmed that he did not delete the Jupyter notebooks from his personal Larkspur Technologies account after leaving Lumos.

When asked to explain why he failed to comply with the return-of-materials obligation in the PIIAA, Yun's response was strikingly casual. He stated that he "forgot" about the obligation and "didn't think about it" in the context of his departure. He denied that anyone at Lumos reminded him of the obligation at the time of his resignation, though he also acknowledged that he did not participate in an exit interview where such a reminder might have been provided.

The notebooks remained on Yun's personal Larkspur Technologies account at least through the dates reflected in the Brindoval forensic analysis, including as late as November 14, 2022. Yun testified that he had "since deleted" the notebooks but was unable to provide a specific date on which the deletion occurred. He stated that the deletion occurred "sometime after" he became aware of Lumos's allegations, but he could not specify whether this was before or after the filing of the complaint in this action on February 17, 2023.

V. Use of Lumos Materials at NovaSpark

A. Yun's Role at NovaSpark

Yun described his role as Founding Engineer at NovaSpark as involving "building ML models for financial signal generation from alternative data." He confirmed that he was one of the first technical hires at NovaSpark — following only Anand (who served as CEO and lead architect) and Mallick (who served as VP of Engineering). He testified that his core responsibilities at NovaSpark included designing and implementing the ML model architecture, developing feature-engineering pipelines, and conducting model training and evaluation.

Yun confirmed that he worked directly with Anand and Mallick on core product development at NovaSpark from the outset. He described the early NovaSpark team as operating in a highly collaborative, fast-paced startup environment where the three of them constituted the entire technical team for the first several weeks. He acknowledged that NovaSpark's product development began "in earnest" in late October or early November 2022, essentially immediately upon his arrival.

B. Access to and Use of Downloaded Notebooks

Yun's testimony regarding whether he accessed or referenced the 47 Jupyter notebooks after joining NovaSpark represents one of the most significant areas of the deposition for purposes of our misappropriation claims.

When initially asked whether he accessed or referenced any of the 47 Jupyter notebooks after joining NovaSpark, Yun stated flatly that he "did not use any Lumos materials" in his work at NovaSpark. He testified: "I relied on my own skills and general knowledge. I didn't copy anything from Lumos."

However, when confronted with additional forensic evidence from Brindoval Digital Forensics LLC (Exhibit 25), which documented that several of the Jupyter notebook files on his personal Larkspur Technologies account were accessed — specifically opened and, in certain instances, modified — on dates after October 24, 2022 (his NovaSpark start date), Yun's testimony became markedly less definitive. He shifted from a categorical denial to a qualified statement: "It's possible I may have opened some files, but I didn't use them for NovaSpark work."

When shown specific file-access timestamps from Brindoval's analysis, Yun was confronted with evidence that particular notebooks were accessed on the following dates:

• November 2, 2022: Access to notebooks containing feature-engineering pipeline code for LumosEdge;

• November 9, 2022: Access to notebooks containing ensemble model architecture configurations; and

• November 14, 2022: Access to notebooks containing training data schema definitions and hyperparameter tuning results.

Yun acknowledged that each of these dates fell during his employment at NovaSpark. When asked to provide an alternative explanation for why the files were accessed on those dates — such as personal curiosity, accidental opening, or any other non-work-related reason — Yun was unable to offer a coherent or credible explanation. He stated only that he "couldn't recall specifically" why the files were accessed but reiterated that he "didn't copy-paste code" or "directly incorporate" Lumos materials into NovaSpark's codebase.

Defense counsel objected at several points during this line of questioning to the characterization of file "access" as file "use." Ms. Whitfield instructed Yun to distinguish between the two concepts in his answers, arguing that opening a file is not the same as using its contents. While this distinction may be argued at summary judgment or trial, the fact remains that the forensic evidence establishes that Yun accessed Lumos proprietary materials during the precise period when he was actively building NovaSpark's competing product, and Yun was unable to provide any innocent explanation for this access.

C. NovaSpark Product Development and Parallels to LumosEdge

Yun was questioned about the feature-engineering approach used in NovaSpark's product, with a particular focus on parallels to LumosEdge. He acknowledged that NovaSpark's early-stage product used "similar types of financial features" as LumosEdge but was quick to attribute this to "industry-standard approaches." He testified: "These are common techniques in quantitative finance. Anyone working in this area would use similar features."

When asked whether NovaSpark's model architecture bore similarities to the LumosEdge ensemble ML model architecture, Yun adopted a different tactic. He stated that he was "not in a position to compare" because he "didn't memorize LumosEdge's architecture." This assertion was challenged immediately. Counsel pointed out that Yun had served as an ML Engineer II at Lumos for nearly two years (January 2021 through October 2022), that he worked directly on the LumosEdge ensemble model architecture on a daily basis, that he had described his own role as including model development and hyperparameter tuning for that very architecture, and that his access to the architecture was both extensive and recent. Yun did not retract his statement but was visibly uncomfortable and paused at length before reaffirming it.

Yun was also questioned about the timeline of NovaSpark's product development. He testified that NovaSpark had a "functional demo" of its product by approximately January 2023 — roughly three months after the three individual defendants joined the company and commenced product development. When asked whether he considered this timeline to be rapid, Yun stated that "a motivated team can move quickly" but acknowledged that building an ML-driven analytics platform "from scratch" would typically take significantly longer.

This rapid development timeline is highly significant. It is consistent with the expert analysis prepared by Naveen Kapoor and Meridian Cybersecurity Advisors Inc., which concluded that NovaSpark's development pace was consistent with access to and use of Lumos's proprietary technology, and that NovaSpark saved approximately 14 to 18 months of development time by leveraging such technology rather than building its systems independently. The three-month timeline to a functional demo — for a product encompassing feature engineering, model training, data pipeline integration, and a user-facing analytics interface — is, in the expert's opinion, not achievable without substantial reliance on pre-existing proprietary work product.

VI. Communications with Anand and Mallick

A. Pre-Departure Communications

Yun confirmed that he communicated with both Rajiv Anand and Serena Mallick about the NovaSpark opportunity before resigning from Lumos. He testified that these communications took place over a period of approximately five to six weeks, from mid-September 2022 through his resignation on October 21, 2022.

Yun testified that the primary mode of communication among the three was personal text messages and phone calls. He also acknowledged that some communications occurred via personal email accounts (i.e., email accounts not associated with Lumos). He stated that he did not use Lumos email or Lumos Meridian Solutions to discuss the NovaSpark opportunity, and that the use of personal channels was "just how we communicated outside of work."

When asked whether any of these pre-departure communications discussed the taking of Lumos proprietary materials to NovaSpark, Yun denied it. He stated that the discussions focused primarily on "the business opportunity and compensation" — specifically, equity allocations, roles, and the general business plan. He testified that Anand described the NovaSpark vision to him and that he found the opportunity compelling. He denied that there was any discussion of downloading, transferring, or otherwise taking Lumos proprietary materials for use at NovaSpark.

However, Yun acknowledged a significant fact: he did not inform Lumos management that Anand and Mallick had approached him about joining a competing venture while he was still employed at Lumos. He agreed that he continued to perform his duties at Lumos, with access to proprietary systems and data, while simultaneously planning to join a direct competitor. He did not dispute that this created a potential conflict of interest, though he characterized his continued work at Lumos during this period as having been performed "in good faith."

B. Post-Departure Coordination

Yun confirmed that he engaged in regular technical collaboration with Anand and Mallick at NovaSpark beginning on October 24, 2022, his first day of employment. He described an intensive period of product development during which the three founders worked closely together in NovaSpark's Bluxome Street office.

Yun testified that the technical discussions among the three founding team members included "feature engineering strategies," "data pipeline architecture," and "model design decisions." He described these as the core technical workstreams that the team pursued in the initial weeks and months of NovaSpark's operations.

When asked whether these technical discussions drew upon knowledge of Lumos's proprietary methods, systems, or trade secrets, Yun's response was equivocal. He stated: "We relied on our general expertise. All three of us have years of experience in machine learning and financial data. We were drawing on everything we knew." When pressed on whether "everything we knew" included knowledge acquired at Lumos — including knowledge of Lumos's proprietary feature-engineering library, ensemble model architecture, hyperparameter configurations, and DataStream pipeline — Yun stated that he "couldn't clearly delineate" between his general professional knowledge and knowledge that was specific to Lumos's proprietary systems.

This inability or unwillingness to distinguish between general skill and proprietary Lumos knowledge is particularly significant in the context of trade-secret litigation. It effectively concedes that the NovaSpark team's work was informed, at a minimum, by their collective experience at Lumos, including their intimate familiarity with Lumos's proprietary technology stack.

VII. NovaSpark Funding and Investor Awareness

Yun confirmed that he was aware that NovaSpark received a $12 million seed funding round from Blackpine Capital Partners LP. He identified the managing partner of Blackpine as Franklin Osei. Yun testified that the funding round closed on approximately November 15, 2022, less than one month after the three individual defendants began working at NovaSpark.

Yun testified that he participated in at least one pitch meeting with representatives of Blackpine Capital Partners prior to the closing of the funding round. He described the meeting as focused on NovaSpark's technical capabilities and product roadmap. He confirmed that Anand led the presentation and that he (Yun) was present to speak to the ML engineering aspects of the product.

When asked whether NovaSpark disclosed to Blackpine that its founding technical team had recently departed from a direct competitor — specifically, from Lumos Analytics Inc. — Yun stated that he "believed Rajiv handled investor communications" and was "unsure" what specifically was disclosed to Blackpine regarding the team's backgrounds. He testified that, in his presence, the team's professional experience was described in general terms (e.g., expertise in ML-driven financial analytics) without specific reference to Lumos by name, but he qualified this by stating he was not present for all investor communications.

Yun denied that any Lumos proprietary materials — including the 47 Jupyter notebooks on his personal Larkspur Technologies or any other proprietary documents — were shown to investors during the pitch process. He was definitive on this point.

The circumstances of NovaSpark's rapid funding raise are relevant to our damages analysis. The $12 million seed round, closed within weeks of the defendants' departures from Lumos, suggests that NovaSpark was able to present a compelling technical narrative to sophisticated investors based in significant part on technology and expertise derived from Lumos.

VIII. Credibility Observations and Assessment

Based on my attendance at the deposition and review of the rough transcript, I offer the following observations regarding Mr. Yun's credibility as a witness:

Pattern of denial followed by concession. Yun's testimony followed a consistent and notable pattern: initial denial or claimed lack of recollection, followed by partial concessions when confronted with documentary or forensic evidence. This pattern was most pronounced in his testimony regarding the Jupyter notebook downloads (Section IV.A above) and his post-departure access to the notebooks (Section V.B above). In each instance, Yun's initial testimony was contradicted by objective evidence, requiring him to revise his statements. This pattern is likely to undermine his credibility with a factfinder, particularly if the initial denials and subsequent concessions are presented sequentially at trial.

The "inadvertent" characterization is untenable. Yun's characterization of the October 18, 2022 download of 47 Jupyter notebooks as "inadvertent" is inconsistent with the evidence in several respects. First, the volume of the downloads — 47 separate notebook files — is inconsistent with an accidental or passive sync event. Second, the downloads occurred on October 18, 2022, which was exactly three days before Yun's resignation and while he was apparently already planning to join NovaSpark (having been recruited by Anand weeks earlier). Third, the Brindoval forensic report documents that the files were moved from specific internal directories to the Larkspur Technologies sync folder, rather than being passively synced from a default location. The timing, volume, and deliberate nature of the file transfers strongly suggest intentional exfiltration.

The "forgetting" excuse. Yun's claim that he "forgot" to return Lumos proprietary materials upon his departure and "didn't think about it" is undermined by his acknowledged PIIAA obligations and the broader context of the coordinated departures. The three defendants' resignations were clearly planned in advance, with Anand incorporating NovaSpark in Delaware on October 3, 2022 — before either Yun or Mallick had resigned. In this context of deliberate planning and coordination, the claim that Yun simply forgot about his contractual obligation to return proprietary materials strains credulity.

Post-departure file access contradicts initial testimony. The Brindoval forensic evidence showing file access on November 2, 9, and 14, 2022, directly contradicts Yun's initial categorical statement that he "did not use any Lumos materials" at NovaSpark. His subsequent retreat to "it's possible I may have opened some files" is insufficient to explain the evidence, particularly in the absence of any alternative explanation for why the files were accessed on those specific dates.

Relative seniority and potential direction by Anand. Yun was the least senior and least experienced of the three individual defendants. His role at Lumos — Machine Learning Engineer II — was subordinate to both Anand (Principal Engineer, the most senior technical contributor) and Mallick (Senior Software Engineer). Several aspects of Yun's testimony suggest that he may have acted at Anand's direction or instigation, including: (a) Yun's testimony that Anand approached him about NovaSpark and was the "primary organizer" of the venture; (b) Yun's deference to Anand regarding investor communications; and (c) the sequencing of the departures, with Anand departing first and the others following in order of seniority. This dynamic may be relevant to both liability apportionment and potential settlement negotiations with individual defendants.

Defense counsel activity. Defense counsel Patricia Whitfield was active throughout the deposition, making frequent objections on grounds of scope, foundation, characterization, and form. She instructed Yun not to answer approximately 8 to 10 questions on privilege grounds, specifically invoking the attorney-client privilege with respect to post-litigation communications between Yun and his counsel. While the objections were generally within the bounds of standard deposition practice, the frequency of instructions not to answer may warrant a motion to compel if the withheld testimony is material and the privilege claims are not properly supported.

IX. Key Exhibits Referenced During Deposition

The following exhibits were marked and referenced during the course of Mr. Yun's deposition:

Copies of each exhibit as marked are being maintained in the deposition exhibit binder and have been entered into the document management system under the appropriate Matter Number. Exhibit 25 is particularly critical and should be cross-referenced against NovaSpark's source code production when it is received.

X. Comparison to Prior Deposition Testimony (Anand and Mallick)

Having now completed depositions of all three individual defendants — Rajiv Anand (May 15, 2023), Serena Mallick (June 7, 2023), and Derek Yun (June 22, 2023) — the following comparative observations are warranted:

Consistency and candor. Yun's testimony is broadly consistent with, but notably less forthcoming than, Anand's deposition testimony regarding the coordinated nature of the departures and the founding of NovaSpark. Anand, while still guarded, provided somewhat more detailed testimony about the timeline of NovaSpark's formation and the roles each individual was expected to play. Yun, by contrast, was frequently evasive and required confrontation with forensic evidence to acknowledge facts that Anand had conceded more readily.

Corroboration of Anand's organizing role. Yun's testimony corroborates Mallick's deposition testimony that Anand was the primary organizer of both the departures from Lumos and the founding of NovaSpark. All three defendants consistently describe Anand as the initiator: he conceived of the venture, recruited the other two, handled incorporation, and led investor relations. This corroboration strengthens our theory that Anand orchestrated a coordinated scheme to misappropriate Lumos's proprietary technology.

Pattern of denial under forensic pressure. Yun's pattern of initial denial followed by concession when confronted with forensic evidence mirrors the pattern observed in both Anand's and Mallick's depositions. In Anand's deposition, he initially denied forwarding the LumosEdge 3.0 roadmap to a personal email account before conceding the point when shown forensic evidence. Mallick similarly denied cloning the DataStream repository before acknowledging it under confrontation. The consistency of this behavioral pattern across all three defendants is itself probative and may be highlighted to the factfinder at trial.

Differentiation of exfiltration conduct. While the pattern of conduct is consistent, the specific nature of each defendant's exfiltration differs in a manner consistent with their respective roles and access levels at Lumos:

• Rajiv Anand (Principal Engineer): Forwarded the full 37-page LumosEdge 3.0 product roadmap PDF to a personal email account (r.anand.personal@vaultmail.com), reflecting his senior role and access to strategic product planning documents.

• Serena Mallick (Senior Software Engineer): Cloned the entire DataStream ETL pipeline repository (approximately 780 MB) to a USB drive (serial number WD-4837291), reflecting her role on the data engineering team and primary responsibility for DataStream.

• Derek Yun (Machine Learning Engineer II): Downloaded 47 proprietary Jupyter notebooks to a personal Larkspur Technologies account, reflecting his ML-focused role and primary responsibility for model development and feature engineering.

The role-specific nature of the exfiltration further supports the inference that the departures and data theft were coordinated: each defendant took the materials most closely aligned with their individual expertise, which together would constitute a substantially complete technical foundation for building a competing product.

XI. Recommendations and Next Steps

Based on the foregoing summary and analysis of Mr. Yun's deposition testimony, I respectfully recommend the following:

1. Strengthening of trade-secret and breach-of-contract claims. Yun's deposition testimony, particularly the post-departure file-access evidence documented in Exhibit 25, significantly strengthens Lumos's trade-secret misappropriation claim under the California Uniform Trade Secrets Act, Cal. Civ. Code §§ 3426–3426.11, as well as the breach of contract claim predicated on Yun's violations of the PIIAA. His admissions regarding the downloads, his failure to return materials, and the forensic evidence of post-departure access establish the core elements of both claims as to this defendant.

2. Use of specific admissions regarding file-access dates. I recommend that Yun's admissions regarding file access on November 2, November 9, and November 14, 2022 be used as key evidence of actual use of Lumos trade secrets in NovaSpark's product development. These dates coincide with the critical early development period at NovaSpark and directly link Yun's access to Lumos proprietary materials with NovaSpark's product-building activities.

3. Support for damages model. The rapid NovaSpark prototype timeline — approximately three months from the defendants' arrival to a functional demo in January 2023 — supports the unjust enrichment damages model prepared by Meridian Cybersecurity Advisors Inc. Meridian's analysis estimates that NovaSpark saved approximately 14 to 18 months of development time through access to and use of Lumos's proprietary technology, corresponding to a damages range of approximately $8.7 million to $12.4 million based on estimated engineering costs and opportunity costs of independent development.

4. Incorporation into mediation brief. I recommend incorporating Yun's deposition testimony into the mediation brief for the anticipated mediation session scheduled for September 6, 2023, before the Honorable Eleanor Voss (Ret.) at Tidewater Mediation Group. Yun's testimony — and in particular his pattern of denial followed by forensic confrontation — will be compelling to a mediator and will strengthen Lumos's negotiating position.

5. Additional discovery. I recommend evaluating whether Yun's testimony provides a sufficient basis for additional discovery requests directed at NovaSpark, including: (a) production of NovaSpark's source code repositories and version control logs from October 2022 through January 2023; (b) production of NovaSpark's internal communications (Meridian Solutions, email, and other messaging platforms) referencing Lumos, LumosEdge, or any of the 47 Jupyter notebook filenames; and (c) deposition of a NovaSpark Rule 2025.230 person most qualified regarding the company's product development timeline and the technical inputs used in that development.

6. Coordination with Meridian on code comparison. I suggest coordinating with Naveen Kapoor of Meridian Cybersecurity Advisors Inc. to conduct a detailed analysis comparing the specific content of the 47 Jupyter notebooks (once preserved copies are obtained through the litigation hold) against NovaSpark's produced source code, to establish direct derivation. This analysis would move our case beyond the circumstantial evidence of file access to direct evidence of code reuse or adaptation, which would substantially strengthen both the injunctive relief and damages components of our claims.

I am available to discuss these recommendations and to assist with any follow-up work at your convenience.

XII. Distribution and Confidentiality Note

This memorandum constitutes attorney work product prepared in anticipation of litigation and is subject to the protections of the attorney-client privilege and the work product doctrine. It reflects the mental impressions, conclusions, opinions, and legal theories of counsel for Lumos Analytics Inc.

Distribution of this memorandum is strictly limited to the following individuals:

• Wendy Schultz, Lead Associate, Calderwood & Harkness LLP

• Marcus Delgado, Partner, Calderwood & Harkness LLP

• Catherine Bloom, Partner, Calderwood & Harkness LLP

This memorandum is not to be shared with the client or any third party without the prior review and express approval of Marcus Delgado.

Any inadvertent disclosure of this memorandum does not constitute a waiver of any applicable privilege or protection.

Respectfully submitted,

CALDERWOOD & HARKNESS LLP

By: ________

Name: Omar Siddiqui

Title: Working Associate, intellectual Property

Date: June 27, 2023

Exhibit No.	Description
Exhibit 14	Executed Proprietary Information and Inventions Assignment Agreement (PIIAA) of Derek Yun, dated January 2021
Exhibit 22	Brindoval Digital Forensics LLC, Preliminary Forensic Report, dated May 1, 2023
Exhibit 23	Lumos IT access logs for Derek Yun's user account, covering the period September–October 2022
Exhibit 24	List of 47 Jupyter notebook filenames downloaded to Yun's personal Larkspur Technologies account on October 18, 2022
Exhibit 25	Brindoval Digital Forensics LLC forensic analysis of Yun's personal Larkspur Technologies file-access metadata (post-October 24, 2022 access events)
Exhibit 26	NovaSpark Technologies Inc. organizational chart and employee roster, produced in discovery
Exhibit 27	Yun's resignation email to Lumos Analytics Inc., dated October 21, 2022

PRIVILEGED AND CONFIDENTIAL — ATTORNEY WORK PRODUCT
CALDERWOOD & HARKNESS LLP intellectual Property Litigation Group
INTERNAL MEMORANDUM
Date: June 27, 2023
To: Wendy Schultz, Lead Associate, intellectual Property, Calderwood & Harkness LLP
From: Omar Siddiqui, Working Associate, intellectual Property, Calderwood & Harkness LLP
Re: Deposition Summary — Derek Yun (June 22, 2023)
Client: Lumos Analytics Inc.
Matter: Lumos Analytics Inc. v. NovaSpark Technologies Inc. et al., Case No. 23-CV-004871, Superior Court of the State of California, County of Santa Clara
C&H Matter Number: 1008-00005
I. Introduction and Deposition Logistics
This memorandum summarizes the deposition testimony of Derek Yun, taken on June 22, 2023, at Calderwood & Harkness LLP's San Francisco office located at 555 California Street, 40th Floor, San Francisco, California 94104. Mr. Yun is a former Machine Learning Engineer II at Lumos Analytics Inc. ("Lumos") and is currently employed as Founding Engineer at NovaSpark Technologies Inc. ("NovaSpark"). He is one of three individual defendants in this action. At the time of the deposition, Mr. Yun was 31 years of age.
The deponent was represented by Patricia Whitfield of Whitfield, Raines & Gerber LLP, located at 235 Montgomery Street, Suite 1800, San Francisco, California 94104. Samuel Gerber of the same firm was also present on behalf of the deponent throughout the proceeding.
The examination was conducted by Wendy Schultz on behalf of Lumos. I attended the deposition and took contemporaneous notes, which form the basis for this summary in conjunction with the rough transcript received from the court reporter on June 24, 2023. The final transcript runs approximately 287 pages and covers approximately 6.5 hours of testimony, inclusive of a morning break, a lunch break, and two additional short breaks requested by defense counsel.
Also present on behalf of Lumos was Naveen Kapoor, Senior Consultant at Meridian Cybersecurity Advisors Inc., who served as a technical advisor. Mr. Kapoor did not participate in the questioning but was available for consultation during breaks regarding technical issues arising from the testimony.
A certified court reporter was present and transcribed the proceedings. A videographer was also present; the deposition was recorded on video. The deposition was taken under oath pursuant to the California Code of Civil Procedure, and Mr. Yun was duly sworn before examination commenced.
This memorandum is organized thematically rather than in strict chronological order of the questioning, to facilitate review and use of the testimony in connection with upcoming motion practice, the mediation brief, and trial preparation.
II. Background and Employment at Lumos Analytics Inc.
A. Hire Date and Role
Mr. Yun confirmed that he joined Lumos Analytics Inc. in January 2021 as a Machine Learning Engineer II. He testified that his hiring process involved a multi-stage technical interview, including a take-home coding exercise, two live technical interviews, and a behavioral interview. He received an offer letter, which he accepted, and his first day of employment was January 11, 2021.
At Lumos, Yun worked within the ML Infrastructure team. He reported directly to a team lead, with his reporting chain extending ultimately to Rachel Fong, Vice President of Engineering. Yun testified that he worked closely on a day-to-day basis with Rajiv Anand, who held the title of Principal Engineer — ML Infrastructure and who served as the technical lead for the team. Yun described Anand as the most senior technical contributor on the ML Infrastructure team and indicated that he looked to Anand for guidance on architectural and design decisions.
When asked to describe his role and responsibilities, Yun stated that his work consisted primarily of developing and refining machine-learning models for the LumosEdge platform. Specifically, he described his core responsibilities as including feature engineering (the process of creating derived variables from raw data inputs to improve model performance), model training (the iterative process of running ML algorithms against training data), hyperparameter tuning (the optimization of model parameters that govern the training process itself), and model validation (the testing and evaluation of model outputs against held-out data to assess performance and generalizability). He testified that his day-to-day work involved extensive use of Python, TensorFlow, PyTorch, and Jupyter notebooks as primary development tools.
Yun acknowledged that the LumosEdge platform was Lumos's core commercial product and described it as an AI-driven analytics platform designed to generate financial trading signals from alternative data sources for use by hedge funds, asset managers, and proprietary trading desks.
B. Access to Proprietary Materials
Yun's testimony regarding his access to Lumos proprietary materials was detailed and is of particular significance to our trade-secret misappropriation claims. The following key admissions were obtained:
First, Yun acknowledged that he had direct access to Lumos's proprietary training datasets, which he confirmed comprised approximately 14.6 terabytes of historical financial and alternative data. He explained that these datasets included both traditional financial market data and alternative data streams drawn from range non-traditional sources, including satellite imagery analytics, web-scraping outputs, credit card transaction aggregations, and geolocation data. He confirmed that these datasets were compiled and curated by Lumos at substantial cost over several years and were not publicly available.
Second, Yun confirmed that he had access to model hyperparameter configurations for the LumosEdge ensemble ML model architecture. He described the hyperparameter configurations as representing "the result of extensive experimentation by the team, including myself" and acknowledged that these configurations were developed over many months of iterative work. He agreed that the hyperparameter configurations were closely guarded within Lumos and were not shared externally.
Third, Yun confirmed that he had access to Lumos's proprietary feature-engineering libraries, which he acknowledged contained 2,847 individual engineered features for financial signal generation. He described the feature-engineering libraries as "one of the most valuable parts of the LumosEdge platform" because the features represented a cumulative investment of engineering effort in identifying, constructing, and validating derived variables that served as inputs to the ML models. He acknowledged that many of these features were novel and were not derived from publicly available academic literature.
Fourth, Yun acknowledged familiarity with the DataStream ETL (Extract, Transform, Load) pipeline to the extent that it fed data into his ML workflows. He explained that DataStream was the data ingestion and processing system that prepared raw data from Lumos's various vendor integrations for consumption by the ML models. However, Yun stated that his access to the DataStream codebase itself was limited to read-only permissions and that the pipeline was primarily maintained by a separate engineering team. He testified that he interacted with DataStream primarily through its output — the cleaned, normalized, and structured data that arrived in the ML team's development environment.
Fifth, Yun testified that he had "some awareness" of the LumosEdge 3.0 product roadmap through attendance at team meetings and planning sessions. However, he claimed that he never personally possessed the full 37-page LumosEdge 3.0 roadmap document. He stated that roadmap discussions in team meetings tended to focus on ML-related features and timelines, and that he did not have visibility into business development, go-to-market, or pricing aspects of the roadmap. This testimony should be viewed in light of the fact that Rajiv Anand, who is alleged to have forwarded the full 37-page roadmap to a personal email account, was Yun's direct technical lead and primary collaborator.
C. Execution of the PIIAA
Yun confirmed that he executed Lumos's standard Proprietary Information and Inventions Assignment Agreement ("PIIAA") at the time of his hire in January 2021. He testified that the PIIAA was presented to him as part of his onboarding paperwork and that he signed it on or about his first day of employment.
When shown the executed copy of the PIIAA (marked as Exhibit 14 during the deposition), Yun identified his signature on the signature page and acknowledged the document as the agreement he had signed. He confirmed that the document appeared to be a complete and accurate copy of the agreement.
Under questioning, Yun acknowledged that the PIIAA contained the following key provisions:
(a) An assignment of all inventions clause, which assigned to Lumos all rights, title, and interest in any inventions, works of authorship, discoveries, developments, and improvements conceived or reduced to practice during his employment and within the scope of Lumos's business;
(b) A confidentiality and non-disclosure provision surviving termination of employment for a period of three years, pursuant to which Yun agreed not to disclose or use Lumos's proprietary and confidential information;
(c) A non-solicitation of employees clause for a period of 12 months following separation from employment, pursuant to which Yun agreed not to solicit or recruit Lumos employees; and
(d) A covenant not to use or disclose Lumos trade secrets at any time, which by its terms was perpetual and not limited to any post-employment period.
When questioned about his understanding of these restrictive covenant provisions, Yun stated that he "understood in general terms" what the PIIAA required of him. However, he claimed that he did not consult an attorney before signing the agreement and that he did not review the PIIAA again at any point during his employment, including before or after his resignation in October 2022. He characterized his review at the time of signing as "a quick read-through" and stated that he understood the agreement was "standard" and "everyone signed it."
Critically, Yun acknowledged that he did not return all company materials upon termination of his employment, as required by the PIIAA. He further acknowledged that he did not confirm the deletion of Lumos proprietary information from his personal devices or accounts at the time of his departure. The PIIAA contains an express provision requiring the return or certified destruction of all proprietary materials upon separation. Yun's failure to comply with this obligation is directly relevant to both our breach of contract and trade-secret misappropriation claims.
III. Circumstances of Resignation from Lumos
A. Decision to Leave
Yun testified that he first began considering leaving Lumos in approximately August or September 2022. He attributed his initial dissatisfaction to what he described as "limited career growth opportunities" and "a desire to be part of something more entrepreneurial." His testimony on this point was notably vague regarding specific timelines and triggering events.
More significantly, Yun acknowledged that Rajiv Anand approached him in approximately mid-September 2022 about joining a new startup venture, which ultimately became NovaSpark Technologies Inc. Yun testified that Anand described the opportunity in general terms as "building an AI-driven alternative-data analytics platform for hedge funds and trading desks." Yun stated that Anand's pitch appealed to him because of the opportunity to be a founding engineer and because of his respect for Anand's technical abilities.
Yun further testified that he also spoke with Serena Mallick about the NovaSpark opportunity before either of them resigned from Lumos. However, Yun was notably vague on the timing and substance of those conversations with Mallick. When pressed for dates and details, he responded with qualifications such as "I don't remember exactly when" and "we talked about it generally." He was unable or unwilling to specify whether Mallick approached him or he approached Mallick, or whether Anand served as an intermediary between them.
Throughout this portion of the examination, defense counsel Patricia Whitfield was particularly active. She objected to several questions regarding the content of conversations between Yun and Anand on attorney-client privilege grounds, specifically asserting that certain communications occurred after the parties had retained counsel in anticipation of the present litigation. These objections were noted on the record. At Ms. Whitfield's instruction, Yun declined to answer approximately three to four questions in this area. The privilege log implications of these instructions should be evaluated, as several of the withheld communications appear to predate the filing of this action and may not be properly subject to the attorney-client privilege.
B. Resignation
Yun resigned from Lumos on October 21, 2022, effective immediately. He did not provide two weeks' notice or any other advance notice period. He submitted his resignation via email to his direct manager and to Rachel Fong, Vice President of Engineering. When shown his resignation email (Exhibit 27), Yun confirmed its authenticity. The email was brief and provided no reason for his departure beyond a general statement that he had "decided to pursue a new opportunity."
Yun testified that he did not participate in a formal exit interview with Lumos's Human Resources department. He stated that he was not contacted by HR regarding an exit interview and that he did not initiate one. He acknowledged that the employee handbook, with which he was familiar, recommended that departing employees participate in an exit interview process.
Notably, Yun did not disclose at the time of his resignation that he intended to join NovaSpark or any other competing venture. He did not inform Lumos that his former colleague Rajiv Anand had recruited him, nor that he planned to begin work at a direct competitor within days of his departure.
Yun acknowledged that he was aware, at the time he submitted his own resignation, that Anand had already resigned from Lumos on September 30, 2022, and that Mallick had resigned on October 14, 2022. The coordinated timing of the three departures — September 30, October 14, and October 21, 2022, in order of seniority — is significant and will be addressed further in Section X below regarding the comparison with prior deposition testimony.
C. Commencement at NovaSpark
Yun confirmed that he began employment at NovaSpark Technologies Inc. on October 24, 2022, which was three calendar days after he resigned from Lumos and the next business day (October 22 and 23, 2022, fell on a Saturday and Sunday, respectively). His title at NovaSpark was Founding Engineer.
Yun confirmed that NovaSpark was incorporated in the State of Delaware on October 3, 2022, by Rajiv Anand. This incorporation date is notable: it falls only three days after Anand's resignation from Lumos on September 30, 2022, which strongly suggests that the incorporation was planned well in advance of Anand's departure. NovaSpark is headquartered at 88 Bluxome Street, Floor 4, San Francisco, California 94107.
When asked to describe NovaSpark's business, Yun acknowledged that NovaSpark's stated business — AI-driven alternative-data analytics for hedge funds and proprietary trading desks — overlaps directly with Lumos's core LumosEdge offering. Yun attempted to qualify this by stating that "the alternative data analytics space is large and there are many different approaches," but he could not identify any material differentiation between NovaSpark's stated business model and Lumos's when pressed on the point.
IV. The Jupyter Notebook Downloads — Key Exfiltration Testimony
This section addresses the most consequential portion of the deposition: Yun's testimony regarding the downloading of 47 proprietary Jupyter notebooks to his personal Larkspur Technologies account on October 18, 2022, three days before his resignation from Lumos. The testimony in this area followed a pattern of initial denial, incremental concession under confrontation with forensic evidence, and ultimately grudging acknowledgment of the core facts.
A. The October 18, 2022 Downloads
When shown Brindoval Digital Forensics LLC's Preliminary Forensic Report (Exhibit 22, dated May 1, 2023) and the corresponding Lumos IT access logs (Exhibit 23), Yun was questioned in detail about the download of 47 proprietary Jupyter notebooks to his personal Larkspur Technologies account on October 18, 2022.
Yun's testimony evolved significantly over the course of questioning on this topic. Initially, Yun testified that he "did not recall" downloading Jupyter notebooks to a personal Larkspur Technologies account. His exact words, as recorded in the rough transcript, were: "I don't recall doing that, no." When pressed on whether he had a personal Larkspur Technologies account, he acknowledged that he did but stated he "didn't use it for work."
When confronted with the forensic metadata contained in the Brindoval report — including specific timestamps corresponding to the downloads and his Lumos user credentials associated with the file-transfer events — Yun shifted his testimony. He stated that "it is possible" the downloads occurred but that he could not "specifically recall" initiating them.
After further questioning, Yun offered an alternative explanation. He acknowledged that he had a personal Larkspur Technologies account that was synced on his Lumos-issued laptop and stated that files "could have been transferred through that sync." He suggested that the Larkspur Technologies desktop client may have automatically synced files that were present in certain directories on his work machine without his affirmative action.
When this explanation was challenged — given that the Brindoval report documented the files as having been moved from specific Lumos internal directories to the Larkspur Technologies sync folder, rather than having been passively synced — Yun ultimately conceded that he "probably did" download the 47 Jupyter notebooks. However, he characterized the action as "inadvertent" and claimed that he was "backing up personal work notes" rather than deliberately taking proprietary materials. He stated: "I had a lot of notes mixed in with work stuff and I was just trying to save my personal things."
Yun was then shown a detailed list of the 47 notebook filenames (Exhibit 24). The filenames included clear and unmistakable references to Lumos-specific proprietary terms and projects, including but not limited to the following:
•  Filenames referencing "LumosEdge" model configurations;
•  Filenames referencing feature-engineering pipelines with Lumos-specific naming conventions;
•  Filenames containing training data schema definitions tied to Lumos's proprietary data structures;
•  Filenames referencing proprietary ensemble model architectures, including the "LE3" designation associated with the LumosEdge 3.0 development effort; and
•  Filenames containing references to specific data vendor integration outputs processed through the DataStream pipeline.
When confronted with this list, Yun acknowledged that many of the filenames referenced Lumos-specific proprietary terms and project names. He conceded that it would be "obvious from the filenames" that the notebooks contained Lumos-related work. However, he maintained his position that some of the notebooks contained "his own personal notes and experiments" that he considered to be his personal intellectual property, distinct from Lumos's proprietary materials. He was unable to identify which specific notebooks he considered to be personal rather than proprietary.
B. Knowledge of Proprietary Nature
A substantial portion of the deposition was devoted to establishing Yun's knowledge and understanding that the 47 Jupyter notebooks contained proprietary information belonging to Lumos. His testimony on this point, while frequently qualified and hedged, yielded significant admissions.
Yun acknowledged that the Jupyter notebooks were created on Lumos systems, using Lumos computing resources (including Lumos's cloud-hosted development environment and GPU instances), during the course of his employment at Lumos and within the scope of his assigned duties. He agreed that, as a general matter, work product created using company resources during the course of employment belongs to the employer.
Yun further acknowledged that the notebooks contained feature-engineering code developed specifically for the LumosEdge platform. He conceded that the code in the notebooks referenced specific engineered features drawn from the 2,847-feature proprietary library maintained by Lumos's ML Infrastructure team. When asked whether those features were publicly available, Yun stated that "some of the underlying concepts are discussed in academic literature" but admitted that "the specific implementations and the specific features as engineered for LumosEdge were developed internally."
Yun admitted that the notebooks contained model hyperparameter configurations for the LumosEdge ensemble model. He described these configurations as representing "the result of extensive experimentation by the team, including myself" and acknowledged that they embodied significant computational effort and engineering judgment that was not replicable from public sources.
Yun further conceded that the notebooks referenced proprietary training data schemas — the structural definitions and metadata describing how Lumos organized and formatted its 14.6-terabyte training dataset — and that the notebooks contained references to data sourced from Lumos's 43 alternative-data vendor integrations processed through the DataStream ETL pipeline. He acknowledged that the identity of Lumos's data vendors, the specific data feeds obtained from those vendors, and the schemas used to process that data were confidential business information.
Despite these concessions, Yun repeatedly attempted to characterize portions of the work contained in the notebooks as "his own intellectual contribution" and suggested that some of the methods employed were based on "publicly available research papers." He referenced general concepts such as gradient boosting, attention mechanisms, and ensemble stacking as examples of publicly known ML techniques that he used in the notebooks. When counsel pointed out that the application and specific configuration of those techniques for financial signal generation from alternative data — the particular combination and optimization reflected in the notebooks — constituted proprietary Lumos work product, Yun did not directly dispute this characterization but restated his position that "there was personal work mixed in."
Critically, when asked directly whether any of the 47 downloaded Jupyter notebooks consisted entirely of publicly available material with no proprietary Lumos content, Yun could not identify a single one. He paused for an extended period before responding that he "would need to review each one individually" to make that determination, but he was unable to name even one notebook that he was confident contained no Lumos proprietary information.
C. Failure to Return or Delete Materials
Yun confirmed that he did not return the 47 Jupyter notebooks or any other Lumos materials upon his resignation on October 21, 2022. He acknowledged that the PIIAA he signed in January 2021 contained an express provision requiring the return of all company materials, including electronic files and copies thereof, upon termination of employment for any reason. He further confirmed that he did not delete the Jupyter notebooks from his personal Larkspur Technologies account after leaving Lumos.
When asked to explain why he failed to comply with the return-of-materials obligation in the PIIAA, Yun's response was strikingly casual. He stated that he "forgot" about the obligation and "didn't think about it" in the context of his departure. He denied that anyone at Lumos reminded him of the obligation at the time of his resignation, though he also acknowledged that he did not participate in an exit interview where such a reminder might have been provided.
The notebooks remained on Yun's personal Larkspur Technologies account at least through the dates reflected in the Brindoval forensic analysis, including as late as November 14, 2022. Yun testified that he had "since deleted" the notebooks but was unable to provide a specific date on which the deletion occurred. He stated that the deletion occurred "sometime after" he became aware of Lumos's allegations, but he could not specify whether this was before or after the filing of the complaint in this action on February 17, 2023.
V. Use of Lumos Materials at NovaSpark
A. Yun's Role at NovaSpark
Yun described his role as Founding Engineer at NovaSpark as involving "building ML models for financial signal generation from alternative data." He confirmed that he was one of the first technical hires at NovaSpark — following only Anand (who served as CEO and lead architect) and Mallick (who served as VP of Engineering). He testified that his core responsibilities at NovaSpark included designing and implementing the ML model architecture, developing feature-engineering pipelines, and conducting model training and evaluation.
Yun confirmed that he worked directly with Anand and Mallick on core product development at NovaSpark from the outset. He described the early NovaSpark team as operating in a highly collaborative, fast-paced startup environment where the three of them constituted the entire technical team for the first several weeks. He acknowledged that NovaSpark's product development began "in earnest" in late October or early November 2022, essentially immediately upon his arrival.
B. Access to and Use of Downloaded Notebooks
Yun's testimony regarding whether he accessed or referenced the 47 Jupyter notebooks after joining NovaSpark represents one of the most significant areas of the deposition for purposes of our misappropriation claims.
When initially asked whether he accessed or referenced any of the 47 Jupyter notebooks after joining NovaSpark, Yun stated flatly that he "did not use any Lumos materials" in his work at NovaSpark. He testified: "I relied on my own skills and general knowledge. I didn't copy anything from Lumos."
However, when confronted with additional forensic evidence from Brindoval Digital Forensics LLC (Exhibit 25), which documented that several of the Jupyter notebook files on his personal Larkspur Technologies account were accessed — specifically opened and, in certain instances, modified — on dates after October 24, 2022 (his NovaSpark start date), Yun's testimony became markedly less definitive. He shifted from a categorical denial to a qualified statement: "It's possible I may have opened some files, but I didn't use them for NovaSpark work."
When shown specific file-access timestamps from Brindoval's analysis, Yun was confronted with evidence that particular notebooks were accessed on the following dates:
•  November 2, 2022: Access to notebooks containing feature-engineering pipeline code for LumosEdge;
•  November 9, 2022: Access to notebooks containing ensemble model architecture configurations; and
•  November 14, 2022: Access to notebooks containing training data schema definitions and hyperparameter tuning results.
Yun acknowledged that each of these dates fell during his employment at NovaSpark. When asked to provide an alternative explanation for why the files were accessed on those dates — such as personal curiosity, accidental opening, or any other non-work-related reason — Yun was unable to offer a coherent or credible explanation. He stated only that he "couldn't recall specifically" why the files were accessed but reiterated that he "didn't copy-paste code" or "directly incorporate" Lumos materials into NovaSpark's codebase.
Defense counsel objected at several points during this line of questioning to the characterization of file "access" as file "use." Ms. Whitfield instructed Yun to distinguish between the two concepts in his answers, arguing that opening a file is not the same as using its contents. While this distinction may be argued at summary judgment or trial, the fact remains that the forensic evidence establishes that Yun accessed Lumos proprietary materials during the precise period when he was actively building NovaSpark's competing product, and Yun was unable to provide any innocent explanation for this access.
C. NovaSpark Product Development and Parallels to LumosEdge
Yun was questioned about the feature-engineering approach used in NovaSpark's product, with a particular focus on parallels to LumosEdge. He acknowledged that NovaSpark's early-stage product used "similar types of financial features" as LumosEdge but was quick to attribute this to "industry-standard approaches." He testified: "These are common techniques in quantitative finance. Anyone working in this area would use similar features."
When asked whether NovaSpark's model architecture bore similarities to the LumosEdge ensemble ML model architecture, Yun adopted a different tactic. He stated that he was "not in a position to compare" because he "didn't memorize LumosEdge's architecture." This assertion was challenged immediately. Counsel pointed out that Yun had served as an ML Engineer II at Lumos for nearly two years (January 2021 through October 2022), that he worked directly on the LumosEdge ensemble model architecture on a daily basis, that he had described his own role as including model development and hyperparameter tuning for that very architecture, and that his access to the architecture was both extensive and recent. Yun did not retract his statement but was visibly uncomfortable and paused at length before reaffirming it.
Yun was also questioned about the timeline of NovaSpark's product development. He testified that NovaSpark had a "functional demo" of its product by approximately January 2023 — roughly three months after the three individual defendants joined the company and commenced product development. When asked whether he considered this timeline to be rapid, Yun stated that "a motivated team can move quickly" but acknowledged that building an ML-driven analytics platform "from scratch" would typically take significantly longer.
This rapid development timeline is highly significant. It is consistent with the expert analysis prepared by Naveen Kapoor and Meridian Cybersecurity Advisors Inc., which concluded that NovaSpark's development pace was consistent with access to and use of Lumos's proprietary technology, and that NovaSpark saved approximately 14 to 18 months of development time by leveraging such technology rather than building its systems independently. The three-month timeline to a functional demo — for a product encompassing feature engineering, model training, data pipeline integration, and a user-facing analytics interface — is, in the expert's opinion, not achievable without substantial reliance on pre-existing proprietary work product.
VI. Communications with Anand and Mallick
A. Pre-Departure Communications
Yun confirmed that he communicated with both Rajiv Anand and Serena Mallick about the NovaSpark opportunity before resigning from Lumos. He testified that these communications took place over a period of approximately five to six weeks, from mid-September 2022 through his resignation on October 21, 2022.
Yun testified that the primary mode of communication among the three was personal text messages and phone calls. He also acknowledged that some communications occurred via personal email accounts (i.e., email accounts not associated with Lumos). He stated that he did not use Lumos email or Lumos Meridian Solutions to discuss the NovaSpark opportunity, and that the use of personal channels was "just how we communicated outside of work."
When asked whether any of these pre-departure communications discussed the taking of Lumos proprietary materials to NovaSpark, Yun denied it. He stated that the discussions focused primarily on "the business opportunity and compensation" — specifically, equity allocations, roles, and the general business plan. He testified that Anand described the NovaSpark vision to him and that he found the opportunity compelling. He denied that there was any discussion of downloading, transferring, or otherwise taking Lumos proprietary materials for use at NovaSpark.
However, Yun acknowledged a significant fact: he did not inform Lumos management that Anand and Mallick had approached him about joining a competing venture while he was still employed at Lumos. He agreed that he continued to perform his duties at Lumos, with access to proprietary systems and data, while simultaneously planning to join a direct competitor. He did not dispute that this created a potential conflict of interest, though he characterized his continued work at Lumos during this period as having been performed "in good faith."
B. Post-Departure Coordination
Yun confirmed that he engaged in regular technical collaboration with Anand and Mallick at NovaSpark beginning on October 24, 2022, his first day of employment. He described an intensive period of product development during which the three founders worked closely together in NovaSpark's Bluxome Street office.
Yun testified that the technical discussions among the three founding team members included "feature engineering strategies," "data pipeline architecture," and "model design decisions." He described these as the core technical workstreams that the team pursued in the initial weeks and months of NovaSpark's operations.
When asked whether these technical discussions drew upon knowledge of Lumos's proprietary methods, systems, or trade secrets, Yun's response was equivocal. He stated: "We relied on our general expertise. All three of us have years of experience in machine learning and financial data. We were drawing on everything we knew." When pressed on whether "everything we knew" included knowledge acquired at Lumos — including knowledge of Lumos's proprietary feature-engineering library, ensemble model architecture, hyperparameter configurations, and DataStream pipeline — Yun stated that he "couldn't clearly delineate" between his general professional knowledge and knowledge that was specific to Lumos's proprietary systems.
This inability or unwillingness to distinguish between general skill and proprietary Lumos knowledge is particularly significant in the context of trade-secret litigation. It effectively concedes that the NovaSpark team's work was informed, at a minimum, by their collective experience at Lumos, including their intimate familiarity with Lumos's proprietary technology stack.
VII. NovaSpark Funding and Investor Awareness
Yun confirmed that he was aware that NovaSpark received a $12 million seed funding round from Blackpine Capital Partners LP. He identified the managing partner of Blackpine as Franklin Osei. Yun testified that the funding round closed on approximately November 15, 2022, less than one month after the three individual defendants began working at NovaSpark.
Yun testified that he participated in at least one pitch meeting with representatives of Blackpine Capital Partners prior to the closing of the funding round. He described the meeting as focused on NovaSpark's technical capabilities and product roadmap. He confirmed that Anand led the presentation and that he (Yun) was present to speak to the ML engineering aspects of the product.
When asked whether NovaSpark disclosed to Blackpine that its founding technical team had recently departed from a direct competitor — specifically, from Lumos Analytics Inc. — Yun stated that he "believed Rajiv handled investor communications" and was "unsure" what specifically was disclosed to Blackpine regarding the team's backgrounds. He testified that, in his presence, the team's professional experience was described in general terms (e.g., expertise in ML-driven financial analytics) without specific reference to Lumos by name, but he qualified this by stating he was not present for all investor communications.
Yun denied that any Lumos proprietary materials — including the 47 Jupyter notebooks on his personal Larkspur Technologies or any other proprietary documents — were shown to investors during the pitch process. He was definitive on this point.
The circumstances of NovaSpark's rapid funding raise are relevant to our damages analysis. The $12 million seed round, closed within weeks of the defendants' departures from Lumos, suggests that NovaSpark was able to present a compelling technical narrative to sophisticated investors based in significant part on technology and expertise derived from Lumos.
VIII. Credibility Observations and Assessment
Based on my attendance at the deposition and review of the rough transcript, I offer the following observations regarding Mr. Yun's credibility as a witness:
Pattern of denial followed by concession. Yun's testimony followed a consistent and notable pattern: initial denial or claimed lack of recollection, followed by partial concessions when confronted with documentary or forensic evidence. This pattern was most pronounced in his testimony regarding the Jupyter notebook downloads (Section IV.A above) and his post-departure access to the notebooks (Section V.B above). In each instance, Yun's initial testimony was contradicted by objective evidence, requiring him to revise his statements. This pattern is likely to undermine his credibility with a factfinder, particularly if the initial denials and subsequent concessions are presented sequentially at trial.
The "inadvertent" characterization is untenable. Yun's characterization of the October 18, 2022 download of 47 Jupyter notebooks as "inadvertent" is inconsistent with the evidence in several respects. First, the volume of the downloads — 47 separate notebook files — is inconsistent with an accidental or passive sync event. Second, the downloads occurred on October 18, 2022, which was exactly three days before Yun's resignation and while he was apparently already planning to join NovaSpark (having been recruited by Anand weeks earlier). Third, the Brindoval forensic report documents that the files were moved from specific internal directories to the Larkspur Technologies sync folder, rather than being passively synced from a default location. The timing, volume, and deliberate nature of the file transfers strongly suggest intentional exfiltration.
The "forgetting" excuse. Yun's claim that he "forgot" to return Lumos proprietary materials upon his departure and "didn't think about it" is undermined by his acknowledged PIIAA obligations and the broader context of the coordinated departures. The three defendants' resignations were clearly planned in advance, with Anand incorporating NovaSpark in Delaware on October 3, 2022 — before either Yun or Mallick had resigned. In this context of deliberate planning and coordination, the claim that Yun simply forgot about his contractual obligation to return proprietary materials strains credulity.
Post-departure file access contradicts initial testimony. The Brindoval forensic evidence showing file access on November 2, 9, and 14, 2022, directly contradicts Yun's initial categorical statement that he "did not use any Lumos materials" at NovaSpark. His subsequent retreat to "it's possible I may have opened some files" is insufficient to explain the evidence, particularly in the absence of any alternative explanation for why the files were accessed on those specific dates.
Relative seniority and potential direction by Anand. Yun was the least senior and least experienced of the three individual defendants. His role at Lumos — Machine Learning Engineer II — was subordinate to both Anand (Principal Engineer, the most senior technical contributor) and Mallick (Senior Software Engineer). Several aspects of Yun's testimony suggest that he may have acted at Anand's direction or instigation, including: (a) Yun's testimony that Anand approached him about NovaSpark and was the "primary organizer" of the venture; (b) Yun's deference to Anand regarding investor communications; and (c) the sequencing of the departures, with Anand departing first and the others following in order of seniority. This dynamic may be relevant to both liability apportionment and potential settlement negotiations with individual defendants.
Defense counsel activity. Defense counsel Patricia Whitfield was active throughout the deposition, making frequent objections on grounds of scope, foundation, characterization, and form. She instructed Yun not to answer approximately 8 to 10 questions on privilege grounds, specifically invoking the attorney-client privilege with respect to post-litigation communications between Yun and his counsel. While the objections were generally within the bounds of standard deposition practice, the frequency of instructions not to answer may warrant a motion to compel if the withheld testimony is material and the privilege claims are not properly supported.
IX. Key Exhibits Referenced During Deposition
The following exhibits were marked and referenced during the course of Mr. Yun's deposition:
Copies of each exhibit as marked are being maintained in the deposition exhibit binder and have been entered into the document management system under the appropriate Matter Number. Exhibit 25 is particularly critical and should be cross-referenced against NovaSpark's source code production when it is received.
X. Comparison to Prior Deposition Testimony (Anand and Mallick)
Having now completed depositions of all three individual defendants — Rajiv Anand (May 15, 2023), Serena Mallick (June 7, 2023), and Derek Yun (June 22, 2023) — the following comparative observations are warranted:
Consistency and candor. Yun's testimony is broadly consistent with, but notably less forthcoming than, Anand's deposition testimony regarding the coordinated nature of the departures and the founding of NovaSpark. Anand, while still guarded, provided somewhat more detailed testimony about the timeline of NovaSpark's formation and the roles each individual was expected to play. Yun, by contrast, was frequently evasive and required confrontation with forensic evidence to acknowledge facts that Anand had conceded more readily.
Corroboration of Anand's organizing role. Yun's testimony corroborates Mallick's deposition testimony that Anand was the primary organizer of both the departures from Lumos and the founding of NovaSpark. All three defendants consistently describe Anand as the initiator: he conceived of the venture, recruited the other two, handled incorporation, and led investor relations. This corroboration strengthens our theory that Anand orchestrated a coordinated scheme to misappropriate Lumos's proprietary technology.
Pattern of denial under forensic pressure. Yun's pattern of initial denial followed by concession when confronted with forensic evidence mirrors the pattern observed in both Anand's and Mallick's depositions. In Anand's deposition, he initially denied forwarding the LumosEdge 3.0 roadmap to a personal email account before conceding the point when shown forensic evidence. Mallick similarly denied cloning the DataStream repository before acknowledging it under confrontation. The consistency of this behavioral pattern across all three defendants is itself probative and may be highlighted to the factfinder at trial.
Differentiation of exfiltration conduct. While the pattern of conduct is consistent, the specific nature of each defendant's exfiltration differs in a manner consistent with their respective roles and access levels at Lumos:
•  Rajiv Anand (Principal Engineer): Forwarded the full 37-page LumosEdge 3.0 product roadmap PDF to a personal email account (r.anand.personal@vaultmail.com), reflecting his senior role and access to strategic product planning documents.
•  Serena Mallick (Senior Software Engineer): Cloned the entire DataStream ETL pipeline repository (approximately 780 MB) to a USB drive (serial number WD-4837291), reflecting her role on the data engineering team and primary responsibility for DataStream.
•  Derek Yun (Machine Learning Engineer II): Downloaded 47 proprietary Jupyter notebooks to a personal Larkspur Technologies account, reflecting his ML-focused role and primary responsibility for model development and feature engineering.
The role-specific nature of the exfiltration further supports the inference that the departures and data theft were coordinated: each defendant took the materials most closely aligned with their individual expertise, which together would constitute a substantially complete technical foundation for building a competing product.
XI. Recommendations and Next Steps
Based on the foregoing summary and analysis of Mr. Yun's deposition testimony, I respectfully recommend the following:
1.  Strengthening of trade-secret and breach-of-contract claims. Yun's deposition testimony, particularly the post-departure file-access evidence documented in Exhibit 25, significantly strengthens Lumos's trade-secret misappropriation claim under the California Uniform Trade Secrets Act, Cal. Civ. Code §§ 3426–3426.11, as well as the breach of contract claim predicated on Yun's violations of the PIIAA. His admissions regarding the downloads, his failure to return materials, and the forensic evidence of post-departure access establish the core elements of both claims as to this defendant.
2.  Use of specific admissions regarding file-access dates. I recommend that Yun's admissions regarding file access on November 2, November 9, and November 14, 2022 be used as key evidence of actual use of Lumos trade secrets in NovaSpark's product development. These dates coincide with the critical early development period at NovaSpark and directly link Yun's access to Lumos proprietary materials with NovaSpark's product-building activities.
3.  Support for damages model. The rapid NovaSpark prototype timeline — approximately three months from the defendants' arrival to a functional demo in January 2023 — supports the unjust enrichment damages model prepared by Meridian Cybersecurity Advisors Inc. Meridian's analysis estimates that NovaSpark saved approximately 14 to 18 months of development time through access to and use of Lumos's proprietary technology, corresponding to a damages range of approximately $8.7 million to $12.4 million based on estimated engineering costs and opportunity costs of independent development.
4.  Incorporation into mediation brief. I recommend incorporating Yun's deposition testimony into the mediation brief for the anticipated mediation session scheduled for September 6, 2023, before the Honorable Eleanor Voss (Ret.) at Tidewater Mediation Group. Yun's testimony — and in particular his pattern of denial followed by forensic confrontation — will be compelling to a mediator and will strengthen Lumos's negotiating position.
5.  Additional discovery. I recommend evaluating whether Yun's testimony provides a sufficient basis for additional discovery requests directed at NovaSpark, including: (a) production of NovaSpark's source code repositories and version control logs from October 2022 through January 2023; (b) production of NovaSpark's internal communications (Meridian Solutions, email, and other messaging platforms) referencing Lumos, LumosEdge, or any of the 47 Jupyter notebook filenames; and (c) deposition of a NovaSpark Rule 2025.230 person most qualified regarding the company's product development timeline and the technical inputs used in that development.
6.  Coordination with Meridian on code comparison. I suggest coordinating with Naveen Kapoor of Meridian Cybersecurity Advisors Inc. to conduct a detailed analysis comparing the specific content of the 47 Jupyter notebooks (once preserved copies are obtained through the litigation hold) against NovaSpark's produced source code, to establish direct derivation. This analysis would move our case beyond the circumstantial evidence of file access to direct evidence of code reuse or adaptation, which would substantially strengthen both the injunctive relief and damages components of our claims.
I am available to discuss these recommendations and to assist with any follow-up work at your convenience.
XII. Distribution and Confidentiality Note
This memorandum constitutes attorney work product prepared in anticipation of litigation and is subject to the protections of the attorney-client privilege and the work product doctrine. It reflects the mental impressions, conclusions, opinions, and legal theories of counsel for Lumos Analytics Inc.
Distribution of this memorandum is strictly limited to the following individuals:
•  Wendy Schultz, Lead Associate, Calderwood & Harkness LLP
•  Marcus Delgado, Partner, Calderwood & Harkness LLP
•  Catherine Bloom, Partner, Calderwood & Harkness LLP
This memorandum is not to be shared with the client or any third party without the prior review and express approval of Marcus Delgado.
Any inadvertent disclosure of this memorandum does not constitute a waiver of any applicable privilege or protection.
Respectfully submitted,
CALDERWOOD & HARKNESS LLP
By: ________
Name: Omar Siddiqui
Title: Working Associate, intellectual Property
Date: June 27, 2023
Exhibit No. | Description
Exhibit 14 | Executed Proprietary Information and Inventions Assignment Agreement (PIIAA) of Derek Yun, dated January 2021
Exhibit 22 | Brindoval Digital Forensics LLC, Preliminary Forensic Report, dated May 1, 2023
Exhibit 23 | Lumos IT access logs for Derek Yun's user account, covering the period September–October 2022
Exhibit 24 | List of 47 Jupyter notebook filenames downloaded to Yun's personal Larkspur Technologies account on October 18, 2022
Exhibit 25 | Brindoval Digital Forensics LLC forensic analysis of Yun's personal Larkspur Technologies file-access metadata (post-October 24, 2022 access events)
Exhibit 26 | NovaSpark Technologies Inc. organizational chart and employee roster, produced in discovery
Exhibit 27 | Yun's resignation email to Lumos Analytics Inc., dated October 21, 2022

PRIVILEGED AND CONFIDENTIAL — ATTORNEY WORK PRODUCT

INTERNAL MEMORANDUM

CALDERWOOD & HARKNESS LLP

Record

Parties

Dates

Amounts