1008-00005 · Lumos Analytics Inc · Pleadings

declaration-chandrasekaran.docx DOCUMENT Litigation

19 / 35 ← → ✕

SUPERIOR COURT OF THE STATE OF CALIFORNIA

COUNTY OF SANTA CLARA

Filed: January 19, 2023

I, Priya Chandrasekaran, hereby declare as follows:

1. I am over the age of eighteen years and am competent to testify to the matters set forth herein. I make this declaration based on my own personal knowledge, except where otherwise stated, and if called as a witness I could and would testify competently to the facts stated in this declaration. I am the Chief Executive Officer of Plaintiff Lumos Analytics Inc. ("Lumos" or "Lumos AI"), a position I have held since the company's founding in approximately 2017. In my capacity as Chief Executive Officer, I have personal knowledge of Lumos's business operations, proprietary technology, personnel decisions, competitive landscape, investor relations, and the facts and circumstances giving rise to the claims asserted in this action. I submit this declaration in support of Lumos's Ex Parte Application for a Temporary Restraining Order and Order to Show Cause Re: Preliminary Injunction.

I. BACKGROUND OF LUMOS ANALYTICS INC.

A. Corporate Formation and Structure

2. Lumos Analytics Inc. is a corporation duly organized and existing under the laws of the State of Delaware, bearing Delaware File Number 6518274 and Employer Identification Number 82-4713956. Lumos maintains its principal place of business at 4700 Orchard Parkway, Suite 300, San Jose, California 95134. I founded Lumos in approximately 2017 together with a small team of colleagues who shared a vision for applying advanced artificial intelligence and machine-learning techniques to the institutional financial analytics market.

3. Lumos is a software-as-a-service ("SaaS") company that specializes in AI-driven financial analytics for institutional asset managers, including hedge funds, pension funds, endowments, and proprietary trading desks. Since its founding, Lumos has grown from a small startup of fewer than ten employees into a substantial enterprise employing approximately 340 people as of the date of this declaration.

B. Products and Market Position

4. Lumos's flagship product is "LumosEdge," a sophisticated platform that uses proprietary ensemble machine-learning models to generate predictive signals for equity and fixed-income portfolio optimization. LumosEdge is designed for and marketed to institutional asset managers who require differentiated, competitively sensitive analytical capabilities that go beyond what is available from commodity analytics providers. LumosEdge processes vast quantities of traditional and alternative financial data to deliver actionable, high-conviction investment signals that our customers rely upon to manage billions of dollars in assets.

5. The proprietary technology underlying LumosEdge represents years of intensive research, development, and significant financial investment. LumosEdge's competitive differentiation rests on the quality and uniqueness of its underlying machine-learning model architecture, its proprietary feature-engineering libraries, its data ingestion and transformation pipelines, and its curated training datasets. These assets, individually and collectively, form the core intellectual property that distinguishes Lumos from its competitors and enables Lumos to charge premium pricing for its services. Without the protection of these proprietary assets, Lumos would be unable to maintain its market position.

C. Business Scale and Financial Position

6. As noted above, Lumos employs approximately 340 people as of the date of this declaration, including software engineers, machine-learning researchers, data scientists, product managers, sales professionals, and support staff. The majority of these employees are located at Lumos's San Jose headquarters, with additional personnel working from satellite offices and remote locations across the United States.

7. Lumos's 2022 annual recurring revenue ("ARR") was approximately $67 million, reflecting the significant value that institutional customers place on Lumos's proprietary analytical capabilities. Our revenue has grown substantially year-over-year since the company's early stages, driven by the differentiated performance of LumosEdge and the expansion of our customer base within the institutional finance sector.

8. To fund the research, development, and commercialization of its proprietary technology, Lumos has raised a total of approximately $185 million in venture capital funding through its Series D round. This capital was raised through successive funding rounds, each reflecting the growing value and promise of Lumos's technology and market opportunity. By way of specific example, Lumos's Series B round of $40 million was led by Ridgeline Ventures in 2019, at a time when LumosEdge was gaining significant market traction among early institutional customers. More recently, Lumos's Series D round of $72 million was led by Northstar Growth Equity and closed in June 2022, just months before the events giving rise to this litigation. The Series D round valued Lumos at a level reflecting the substantial maturity and competitive strength of the company's proprietary technology platform.

9. Lumos's investor base includes prominent venture capital and growth equity firms. Helen Trask, General Partner at Ridgeline Ventures, serves as a board observer for Lumos, and Jonathan Holt, Managing Director at Northstar Growth Equity, serves as a member of Lumos's Board of Directors. These investors committed substantial capital based in significant part on the value and defensibility of Lumos's proprietary technology. The significant capital raised by Lumos is a direct reflection of the substantial value of its proprietary technology assets and its competitive market position — assets that are now threatened by the conduct of Defendants described below.

II. LUMOS'S PROPRIETARY TECHNOLOGY AND TRADE SECRETS

A. Overview of Trade Secret Categories

10. Over the course of approximately six years, Lumos has developed, at great expense and through the dedicated efforts of its engineering and research teams, proprietary technology and confidential business information that constitutes its most valuable competitive assets. These assets are not publicly known or generally ascertainable, and Lumos has taken extensive and continuous measures to maintain their secrecy. Five categories of trade secrets are at issue in this matter, each of which I describe in detail below.

B. LumosEdge Ensemble Machine-Learning Model Architecture

11. The first category of trade secrets at issue is the LumosEdge ensemble machine-learning model architecture, including its proprietary neural-network topology and hyperparameter configurations. This architecture is the core intellectual engine of Lumos's entire product offering. It defines how LumosEdge combines multiple machine-learning models — including deep neural networks, gradient-boosted decision trees, and proprietary attention-based architectures — into an ensemble framework that produces predictive financial signals of exceptional accuracy and robustness.

12. The model architecture was developed through extensive research and development investment over a period of years, drawing on the specialized expertise of some of the most talented machine-learning engineers and researchers in the industry. The specific neural-network topologies, the manner in which individual model components are combined, and the hyperparameter configurations that govern the training and inference processes are the product of thousands of hours of experimentation, iteration, and refinement. This architecture is not publicly known or generally available in the field. Access to the model architecture specifications is strictly limited to authorized Lumos personnel who have a demonstrated need to know and who have executed appropriate confidentiality agreements.

C. Proprietary Feature-Engineering Libraries

13. The second category of trade secrets at issue is Lumos's proprietary feature-engineering libraries, which contain 2,847 individual engineered features used for financial signal generation. Feature engineering — the process of transforming raw data inputs into informative variables that enhance the predictive power of machine-learning models — is widely recognized as one of the most critical determinants of model performance in applied machine learning.

14. Lumos's feature-engineering libraries were developed iteratively over years of painstaking research and experimentation. Each of the 2,847 features represents a carefully designed transformation of one or more data inputs that has been empirically validated to contribute meaningfully to the predictive accuracy of LumosEdge's models. Many of these features encode proprietary insights about the relationships between alternative data sources and financial market behavior — insights that were arrived at only after extensive trial and error and that are not obvious or derivable from publicly available information. The library of engineered features represents a significant competitive moat. It cannot be independently replicated by a competitor without an investment of substantial time, capital, and expertise comparable to what Lumos itself expended to create it.

D. DataStream ETL Pipeline

15. The third category of trade secrets at issue is the "DataStream" ETL (extract, transform, load) pipeline codebase and associated data-schema definitions. DataStream is Lumos's proprietary system for ingesting, transforming, and loading data from a diverse array of sources into the LumosEdge analytical framework. DataStream covers ingestion from forty-three (43) separate alternative-data vendors, each of which provides data in different formats, with different delivery mechanisms, and subject to different quality and reliability characteristics.

16. The DataStream pipeline represents a proprietary system that enables Lumos to integrate disparate and complex data sources — including satellite imagery analytics, credit card transaction data, web traffic data, geolocation data, and many others — into a unified analytical framework suitable for consumption by LumosEdge's machine-learning models. The pipeline architecture, schema definitions, vendor integration logic, data-quality validation rules, and error-handling procedures are all confidential and proprietary to Lumos. Building a comparable ETL pipeline from scratch, including negotiating data access agreements with dozens of alternative-data vendors and developing the specialized integration logic for each, would require many months if not years of effort and millions of dollars in investment.

E. Curated Training Dataset

17. The fourth category of trade secrets at issue is a curated and labeled training dataset comprising approximately 14.6 terabytes of historical financial and alternative data. This dataset was assembled, cleaned, labeled, and curated at great expense and over a significant period of time. It includes historical market data, alternative data from dozens of vendors, proprietary labels reflecting actual financial outcomes, and metadata generated during the feature-engineering process.

18. The curated training dataset is foundational to the performance of LumosEdge's machine-learning models. The quality and comprehensiveness of training data is a critical determinant of model accuracy, and the extensive curation effort that Lumos invested in this dataset — including the removal of noise, the correction of errors, the imputation of missing values, and the careful alignment of disparate time series — is a core component of Lumos's competitive advantage. The dataset is stored on secured internal infrastructure with strict access controls, including role-based permissions, audit logging, and encryption at rest and in transit.

F. Confidential Product Roadmap

19. The fifth category of trade secrets at issue is Lumos's confidential product roadmap documents for the period 2023–2025, including, most critically, the unreleased specifications for "LumosEdge 3.0." The LumosEdge 3.0 roadmap document is thirty-seven (37) pages in length and details planned product features, architecture changes, integration plans, go-to-market strategies, and strategic initiatives that Lumos intends to pursue over the next several years.

20. The product roadmap constitutes some of the most competitively sensitive information in Lumos's possession. Disclosure of roadmap information to a direct competitor would cause severe and immediate strategic harm to Lumos by enabling the competitor to anticipate Lumos's product moves, preemptively develop competing features, and undercut Lumos's strategic positioning in the marketplace. The roadmap was distributed only to a limited number of senior Lumos employees and was marked "CONFIDENTIAL — INTERNAL USE ONLY."

G. Protective Measures

21. Lumos takes extensive and continuous measures to protect its trade secrets and confidential information from unauthorized access, use, and disclosure. These measures include, but are not limited to, the following:

(a) Restricting access to proprietary technology and confidential information on a strict need-to-know basis, with role-based access controls enforced through technical systems;

(b) Requiring all employees who have access to proprietary technology and confidential information to execute Lumos's standard Proprietary Information and Inventions Assignment Agreement ("PIIAA") as a condition of their employment;

(c) Employing comprehensive technical access controls and monitoring, including multi-factor authentication, encrypted storage, network segmentation, endpoint detection and response software, and audit logging of access to sensitive repositories and databases;

(d) Maintaining physical security protocols at Lumos's offices, including badge-access entry, visitor registration, and restricted access to server rooms and engineering workspaces; and

(e) Conducting regular security awareness training for all employees and periodic security audits of systems containing proprietary information.

22. These measures, taken together, demonstrate that Lumos treats its proprietary technology and confidential information as trade secrets and takes reasonable steps — indeed, extensive steps — to maintain their secrecy.

III. EMPLOYMENT AND DEPARTURE OF THE INDIVIDUAL DEFENDANTS

A. Rajiv Anand

23. Defendant Rajiv Anand, age 37, was employed by Lumos as Principal Engineer — ML Infrastructure from March 2019 through September 30, 2022. In this role, Anand was one of the key architects of the LumosEdge signal-generation engine. His responsibilities included designing, implementing, and optimizing the core machine-learning infrastructure that powers LumosEdge's ensemble models. He reported directly to Lumos's VP of Engineering, Rachel Fong, and worked closely with senior members of the machine-learning and data-engineering teams.

24. By virtue of his senior technical role, Anand had access to Lumos's most sensitive proprietary codebases and training data. Specifically, Anand had authorized access to the LumosEdge ensemble ML model architecture, including the proprietary neural-network topologies and hyperparameter configurations; the feature-engineering libraries; and the curated training datasets. His access to these materials was essential to the performance of his job duties, and he was trusted with this access precisely because of the confidentiality obligations he had undertaken.

25. Upon his hire in March 2019, Anand executed Lumos's standard PIIAA. The PIIAA includes the following restrictive covenant provisions, among others:

(a) An assignment of all inventions clause, pursuant to which Anand assigned to Lumos all right, title, and interest in any inventions, developments, or intellectual property created during or arising out of his employment at Lumos;

(b) A confidentiality and non-disclosure provision that survives the termination of Anand's employment for a period of three (3) years, during which Anand is prohibited from disclosing, using, or exploiting any Lumos confidential information or trade secrets;

(c) A non-solicitation of employees clause prohibiting Anand from soliciting, recruiting, or inducing any Lumos employee to leave Lumos's employ for a period of twelve (12) months following the date of Anand's separation from Lumos; and

(d) A covenant not to use or disclose Lumos trade secrets at any time, without limitation, in accordance with applicable trade secret law.

26. The PIIAA further requires that, upon the termination of employment for any reason, the departing employee must promptly return to Lumos all company materials, including but not limited to documents, files, data, code, hardware, and any other tangible or electronic materials containing or reflecting Lumos's proprietary information, and must certify in writing that all such materials have been returned and that no copies have been retained.

27. Anand resigned from Lumos on September 30, 2022, effective immediately, providing no advance notice. Critically, upon his departure, Anand did not return company materials and did not confirm the deletion of Lumos proprietary information from his personal devices, accounts, or storage media, as required by his PIIAA. Despite Lumos's subsequent requests, Anand has not complied with these obligations.

B. Serena Mallick

28. Defendant Serena Mallick, age 34, was employed by Lumos as Senior Software Engineer — Data Pipelines from June 2020 through October 14, 2022. In this role, Mallick was responsible for building, maintaining, and optimizing the proprietary DataStream ETL pipelines that ingest alternative financial data from dozens of third-party vendors into the LumosEdge platform. She was a key contributor to the DataStream system and possessed deep technical knowledge of its architecture, codebase, schema definitions, and vendor integration configurations.

29. By virtue of her role, Mallick had extensive access to the DataStream ETL pipeline codebase, its data-schema definitions, and the vendor integration configurations for multiple alternative-data providers. Her access to these materials was necessary for the performance of her job duties and was granted subject to the confidentiality and proprietary information protections described herein.

30. Upon her hire in June 2020, Mallick executed Lumos's standard PIIAA, which contains the same restrictive covenant provisions as those described above in connection with Anand's PIIAA, including the assignment of inventions clause, the confidentiality and non-disclosure provision surviving termination for three years, the non-solicitation of employees clause for twelve months post-separation, and the perpetual covenant not to use or disclose Lumos trade secrets.

31. Mallick resigned from Lumos on October 14, 2022, also without adequate advance notice. Like Anand, Mallick did not return company materials and did not confirm the deletion of Lumos proprietary information from her personal devices, accounts, or storage media upon her departure, in breach of her obligations under the PIIAA.

C. Derek Yun

32. Defendant Derek Yun, age 31, was employed by Lumos as Machine Learning Engineer II from January 2021 through October 21, 2022. In this role, Yun worked on model development, training, and evaluation within the LumosEdge platform. He had direct access to Lumos's proprietary training datasets, model hyperparameters, and feature-engineering libraries. Yun was involved in the day-to-day process of training and refining LumosEdge's machine-learning models and possessed detailed knowledge of the methodologies and data that underpin those models.

33. Upon his hire in January 2021, Yun executed Lumos's standard PIIAA, which contains the same restrictive covenant provisions as those executed by Anand and Mallick, as described above.

34. Yun resigned from Lumos on October 21, 2022, again without adequate advance notice. Like both Anand and Mallick, Yun did not return company materials and did not confirm the deletion of Lumos proprietary information from his personal devices, accounts, or storage media upon his departure, as required by his PIIAA.

D. Coordinated Nature of Departures

35. The departures of Anand, Mallick, and Yun occurred in rapid succession over a period of approximately three weeks — from September 30, 2022 through October 21, 2022. The timing and coordination of these departures, viewed in light of the events described in Sections IV and V below, strongly suggest a pre-planned and deliberate effort by the individual defendants to leave Lumos in a coordinated fashion and to establish a competing venture using Lumos's proprietary technology and trade secrets. I do not believe these near-simultaneous resignations were coincidental. In all my years leading Lumos, I have never experienced a coordinated departure of this nature, involving employees with such extensive access to the company's most sensitive proprietary materials.

IV. FORMATION OF NOVASPARK TECHNOLOGIES INC. AND COMPETITIVE THREAT

A. NovaSpark's Incorporation and Business

36. NovaSpark Technologies Inc. ("NovaSpark") was incorporated in the State of Delaware on October 3, 2022 — just three days after Anand's resignation from Lumos on September 30, 2022. I am informed and believe that the incorporation paperwork for NovaSpark was prepared and filed while Anand was still employed at Lumos or, at the very latest, within days of his departure, further evidencing that the formation of NovaSpark was pre-planned during the period of Anand's employment.

37. NovaSpark is headquartered at 88 Bluxome Street, Floor 4, San Francisco, California 94107. Defendant Rajiv Anand serves as the Chief Executive Officer and co-founder of NovaSpark.

38. Based on publicly available information, including NovaSpark's website and press releases, NovaSpark purports to offer AI-driven alternative-data analytics for hedge funds and proprietary trading desks. This is a market that directly and substantially overlaps with Lumos's core offering. Indeed, the description of NovaSpark's product and target customer base is remarkably similar to the description of LumosEdge and its target market, raising serious questions about the extent to which NovaSpark's product strategy and technology are derived from Lumos's proprietary materials.

B. Individual Defendants' Roles at NovaSpark

39. Defendant Serena Mallick joined NovaSpark as Vice President of Engineering on October 17, 2022 — just three days after her resignation from Lumos on October 14, 2022. The extraordinarily short interval between her departure from Lumos and her assumption of a senior leadership role at NovaSpark confirms that her departure from Lumos was pre-arranged and coordinated with Anand.

40. Defendant Derek Yun joined NovaSpark as a founding engineer on October 24, 2022 — just three days after his resignation from Lumos on October 21, 2022. As with Mallick, the near-instantaneous transition from Lumos to NovaSpark demonstrates that Yun's departure was part of a coordinated plan.

41. All three individual defendants — Anand, Mallick, and Yun — now hold senior positions at NovaSpark, a company that is a direct competitor to Lumos. Together, these three individuals possess deep and comprehensive knowledge of virtually every category of Lumos trade secret at issue in this case: the ML model architecture, the feature-engineering libraries, the DataStream ETL pipeline, the curated training datasets, and the confidential product roadmap. The concentration of this knowledge within a single competing startup is deeply concerning.

C. NovaSpark's Funding

42. On or about November 15, 2022, NovaSpark closed a $12 million seed funding round led by Blackpine Capital Partners LP. Franklin Osei is the Managing Partner of Blackpine Capital Partners LP, which is headquartered at 101 Mission Street, 38th Floor, San Francisco, California 94105. The closing of a $12 million seed round occurred approximately six weeks after NovaSpark's incorporation, at a time when the company was barely a month old and could not have developed significant proprietary technology of its own from scratch in such a brief period.

43. The rapid securing of significant venture capital funding for a startup of such recent vintage raises serious concerns about what assets, capabilities, or proprietary knowledge NovaSpark represented to its investors in order to secure such funding. A reasonable inference is that NovaSpark's pitch to Blackpine Capital Partners included, implicitly or explicitly, representations about the founders' deep expertise and access to technology and data that was in fact Lumos's proprietary information.

D. Direct Competitive Overlap

44. NovaSpark's stated business — AI-driven alternative-data analytics for institutional finance clients — directly competes with Lumos's core LumosEdge offering. Both companies target the same customer base: institutional asset managers, hedge funds, and proprietary trading desks that seek AI-powered analytics to gain a competitive edge in financial markets.

45. Given the departing employees' deep and comprehensive knowledge of Lumos's proprietary technology, NovaSpark's ability to rapidly build a competing product is inextricably linked to the trade secrets the individual defendants had access to during their tenure at Lumos. It strains credulity to believe that Anand, Mallick, and Yun could develop, in a matter of weeks or months, a product comparable to LumosEdge without drawing upon the proprietary knowledge, methodologies, and materials they acquired at Lumos over years of employment.

V. DISCOVERY OF DATA EXFILTRATION

A. IT Security Investigation

46. On December 12, 2022, Lumos's IT security team, in the course of a routine security audit, identified anomalous data-exfiltration activity associated with the corporate accounts and workstations of each of the three individual defendants during the final weeks of their respective employment at Lumos. Upon detecting these anomalies, the IT security team immediately escalated the matter for further investigation.

47. The investigation was conducted in coordination with Rachel Fong, Lumos's VP of Engineering, who possesses the technical expertise to understand the significance and sensitivity of the materials that were exfiltrated. Ms. Fong can attest to the technical details of the investigation and the nature of the compromised materials. The findings of this investigation are summarized below.

B. Specific Exfiltration Events

48. The investigation revealed the following specific acts of unauthorized data exfiltration by each of the individual defendants:

Rajiv Anand:

49. On September 26, 2022 — four days before his resignation from Lumos — Anand transferred approximately 3.2 gigabytes (GB) of files from Lumos's internal code repositories and configuration management systems to a personal Hollowell Health IT account. Lumos's access logs confirm that the transferred files included model architecture specifications, hyperparameter configuration files, and related technical documentation associated with the LumosEdge ensemble ML model. These files contain some of the most sensitive and valuable proprietary information in Lumos's possession, representing the core intellectual property underlying LumosEdge's predictive capabilities.

50. On September 28, 2022 — two days before his resignation from Lumos — Anand accessed and forwarded the confidential thirty-seven (37) page "LumosEdge 3.0 Product Roadmap" document, in PDF format, to his personal email address, r.anand.personal@vaultmail.com. Lumos's email monitoring systems recorded this transmission. The LumosEdge 3.0 Product Roadmap is among the most competitively sensitive documents in Lumos's possession, containing detailed plans for future product features, architectural enhancements, and strategic initiatives covering the period from 2023 through 2025.

Serena Mallick:

51. On October 10, 2022 — four days before her resignation from Lumos — Mallick cloned the entire DataStream ETL pipeline repository from Lumos's internal version-control system to a personal external USB drive. Lumos's endpoint-detection software recorded the connection of the USB drive (identified by serial number WD-4837291) and the transfer of approximately 780 megabytes (MB) of data to the device. The transferred repository contains the complete DataStream ETL pipeline codebase, including source code, data-schema definitions, vendor integration configurations, and related documentation. This codebase represents the entirety of Lumos's proprietary data-ingestion infrastructure.

Derek Yun:

52. On October 18, 2022 — three days before his resignation from Lumos — Yun downloaded forty-seven (47) proprietary Jupyter notebooks from Lumos's internal development environment to a personal Larkspur Technologies account. Lumos's data-loss-prevention systems detected and logged this transfer. The Jupyter notebooks contain feature-engineering code, model-training scripts, experimental results, and associated commentary documenting the rationale behind specific modeling choices. These notebooks encapsulate significant proprietary knowledge about Lumos's feature-engineering methodologies and model-training processes.

C. Failure to Return Materials

53. As noted above, none of the three individual defendants returned company materials or confirmed the deletion of Lumos proprietary information from their personal devices, accounts, or storage media upon their departure from Lumos, as required by the terms of their respective PIIAAs. Despite Lumos's subsequent demands, none of the individual defendants has complied with these obligations as of the date of this declaration.

54. Upon discovering the data-exfiltration events described above, Lumos acted promptly to assess the scope of the breach and to seek legal counsel. On December 15, 2022, Lumos retained the law firm of Calderwood & Harkness LLP to investigate the matter and to advise on potential litigation. Calderwood & Harkness has been engaged in an intensive investigation since that date, the results of which inform this declaration and the accompanying application for emergency relief.

VI. IRREPARABLE HARM TO LUMOS

A. Nature of Harm

55. The continued possession, use, and potential disclosure of Lumos's trade secrets by Defendants causes and threatens to cause irreparable harm to Lumos that cannot be adequately compensated by monetary damages alone. It is well established that trade secrets, once disclosed, lose their status as trade secrets and cannot be "unlearned" or recalled. The unauthorized possession of Lumos's trade secrets by the individual defendants and their use by or on behalf of NovaSpark creates a continuing and escalating risk that these trade secrets will be further disseminated, incorporated into competing products, or otherwise exploited in ways that permanently destroy their value as proprietary assets.

56. The harm that Lumos faces is not speculative. Defendants have already taken Lumos's trade secrets. Defendants are already building a directly competing product. And Defendants already have the financial resources to bring that product to market. The risk of irreparable harm is immediate, concrete, and severe.

B. Competitive Harm

57. NovaSpark is in direct competition with Lumos for the same institutional customers — asset managers, hedge funds, and proprietary trading desks — that constitute Lumos's core revenue base. If NovaSpark uses Lumos's proprietary technology to develop and launch a competing product, Lumos will lose the competitive advantage that it has built over six years and at a cost of hundreds of millions of dollars in investment. Customers who are offered a product built on technology substantially identical to LumosEdge's will have reduced incentive to contract with Lumos, and Lumos will face the impossible task of competing against its own proprietary technology.

58. Moreover, the harm from the loss of trade-secret protection is inherently difficult, if not impossible, to quantify. Once a trade secret is incorporated into a competitor's product, it becomes exceedingly difficult to disentangle the competitor's independently developed technology from the misappropriated trade secrets. Lost customer relationships, diminished pricing power, and eroded market position are harms that defy precise monetary calculation and are therefore irreparable.

C. Strategic and Reputational Harm

59. The disclosure of the LumosEdge 3.0 product roadmap to a direct competitor would undermine Lumos's multi-year strategic product planning. The roadmap reflects Lumos's assessment of market opportunities, its planned technology investments, and its competitive positioning strategy for the period 2023 through 2025. If NovaSpark possesses this roadmap — which Anand forwarded to his personal email account just days before his departure — NovaSpark will be able to anticipate and preempt Lumos's product moves, neutralizing the strategic advantages that Lumos's planned innovations were designed to create.

60. In addition, Lumos's investors — including Ridgeline Ventures and Northstar Growth Equity — have invested a total of $185 million in Lumos based in significant part on the value and defensibility of Lumos's proprietary technology. If Lumos's trade secrets are dissipated through misappropriation, the foundation of that investment will be undermined. Loss of investor confidence could impair Lumos's ability to raise additional capital, attract and retain top talent, and pursue its strategic objectives. Similarly, the confidence of Lumos's institutional customers — who entrust Lumos with their most sensitive portfolio data and rely on LumosEdge for mission-critical investment decisions — could be severely damaged if those customers learn that Lumos's proprietary technology has been compromised. The resulting reputational harm would further compound the irreparable injury to Lumos.

D. Urgency.

61. Every day that Defendants retain and potentially use Lumos's trade secrets compounds the harm to Lumos. NovaSpark is actively building a competing product, and there is no indication that Defendants have ceased using or relying upon Lumos's proprietary materials. NovaSpark has the financial resources — a $12 million seed round — to accelerate its product development, hire additional engineers, and bring a competing product to market in a compressed timeframe.

62. Absent immediate judicial intervention, the damage to Lumos will be irremediable. Each day that passes without the protection of a temporary restraining order increases the risk that Lumos's trade secrets will be further disseminated, permanently incorporated into NovaSpark's products, or disclosed to additional third parties, including NovaSpark's investors, prospective customers, or new hires. Once that occurs, no amount of monetary damages can restore the status quo ante.

VII. NEED FOR EMERGENCY RELIEF

63. Given the nature and sensitivity of the proprietary information taken by Defendants, the direct and immediate competitive threat posed by NovaSpark, the documented evidence of data exfiltration, the individual defendants' refusal to return company materials, and the ongoing and accelerating risk of irreparable harm, Lumos respectfully submits that emergency relief in the form of a temporary restraining order is both necessary and fully justified.

64. Lumos respectfully requests that the Court order Defendants to:

(a) Immediately cease all use, disclosure, dissemination, and exploitation of Lumos's proprietary information and trade secrets, including but not limited to the LumosEdge ensemble ML model architecture, the proprietary feature-engineering libraries, the DataStream ETL pipeline codebase and data-schema definitions, the curated training dataset, and the confidential product roadmap documents;

(b) Immediately preserve all electronic devices, storage media, cloud accounts, email accounts, and any other repositories containing or potentially containing Lumos materials, and refrain from destroying, deleting, altering, moving, or concealing any documents, files, data, code, or other materials related to Lumos's proprietary technology; and

(c) Refrain from soliciting, recruiting, or inducing any current Lumos employee to leave Lumos's employ or to join NovaSpark or any other enterprise affiliated with any Defendant, in accordance with the non-solicitation provisions of the individual defendants' respective PIIAAs.

65. Without such relief, Lumos faces the imminent and concrete prospect of losing trade secrets that are the very foundation of its business and that represent years of intensive development and hundreds of millions of dollars in investment by Lumos and its investors. The balance of hardships overwhelmingly favors the issuance of a temporary restraining order: while Defendants may be temporarily inconvenienced by an order requiring them to refrain from using another company's stolen property, Lumos faces the permanent destruction of its most valuable assets.

66. Lumos is prepared to post a bond in an amount the Court deems appropriate as security for any damages that Defendants may sustain by reason of the temporary restraining order, should the Court ultimately determine that the order was wrongfully issued. Lumos submits, however, that any bond should be modest given the overwhelming evidence of misappropriation and the minimal hardship to Defendants of an order that merely requires them to refrain from using property that is not theirs.

I declare under penalty of perjury under the laws of the State of California that the foregoing is true and correct.

Executed on January 19, 2023, at San Jose, California.

________________________________________

Priya Chandrasekaran Chief Executive Officer Lumos Analytics Inc. 4700 Orchard Parkway, Suite 300 San Jose, California 95134

Prepared by:

CALDERWOOD & HARKNESS LLP

Marcus Delgado, Responsible Partner Catherine Bloom, Billing Partner Wendy Schultz, Lead Associate Omar Siddiqui, Working Associate

555 California Street, 40th Floor San Francisco, California 94104

Attorneys for Plaintiff Lumos Analytics Inc.

C&H Matter Number: 1008-00005

LUMOS ANALYTICS INC., a Delaware corporation,	Case No.: 23-CV-004871
Plaintiff,	DECLARATION OF PRIYA CHANDRASEKARAN IN SUPPORT OF PLAINTIFF'S EX PARTE APPLICATION FOR TEMPORARY RESTRAINING ORDER AND ORDER TO SHOW CAUSE RE: PRELIMINARY INJUNCTION

NOVASPARK TECHNOLOGIES INC., a Delaware corporation; RAJIV ANAND, an individual; SERENA MALLICK, an individual; and DEREK YUN, an individual,

Defendants.

SUPERIOR COURT OF THE STATE OF CALIFORNIA
COUNTY OF SANTA CLARA
Filed: January 19, 2023
I, Priya Chandrasekaran, hereby declare as follows:
1. I am over the age of eighteen years and am competent to testify to the matters set forth herein. I make this declaration based on my own personal knowledge, except where otherwise stated, and if called as a witness I could and would testify competently to the facts stated in this declaration. I am the Chief Executive Officer of Plaintiff Lumos Analytics Inc. ("Lumos" or "Lumos AI"), a position I have held since the company's founding in approximately 2017. In my capacity as Chief Executive Officer, I have personal knowledge of Lumos's business operations, proprietary technology, personnel decisions, competitive landscape, investor relations, and the facts and circumstances giving rise to the claims asserted in this action. I submit this declaration in support of Lumos's Ex Parte Application for a Temporary Restraining Order and Order to Show Cause Re: Preliminary Injunction.
I. BACKGROUND OF LUMOS ANALYTICS INC.
A. Corporate Formation and Structure
2. Lumos Analytics Inc. is a corporation duly organized and existing under the laws of the State of Delaware, bearing Delaware File Number 6518274 and Employer Identification Number 82-4713956. Lumos maintains its principal place of business at 4700 Orchard Parkway, Suite 300, San Jose, California 95134. I founded Lumos in approximately 2017 together with a small team of colleagues who shared a vision for applying advanced artificial intelligence and machine-learning techniques to the institutional financial analytics market.
3. Lumos is a software-as-a-service ("SaaS") company that specializes in AI-driven financial analytics for institutional asset managers, including hedge funds, pension funds, endowments, and proprietary trading desks. Since its founding, Lumos has grown from a small startup of fewer than ten employees into a substantial enterprise employing approximately 340 people as of the date of this declaration.
B. Products and Market Position
4. Lumos's flagship product is "LumosEdge," a sophisticated platform that uses proprietary ensemble machine-learning models to generate predictive signals for equity and fixed-income portfolio optimization. LumosEdge is designed for and marketed to institutional asset managers who require differentiated, competitively sensitive analytical capabilities that go beyond what is available from commodity analytics providers. LumosEdge processes vast quantities of traditional and alternative financial data to deliver actionable, high-conviction investment signals that our customers rely upon to manage billions of dollars in assets.
5. The proprietary technology underlying LumosEdge represents years of intensive research, development, and significant financial investment. LumosEdge's competitive differentiation rests on the quality and uniqueness of its underlying machine-learning model architecture, its proprietary feature-engineering libraries, its data ingestion and transformation pipelines, and its curated training datasets. These assets, individually and collectively, form the core intellectual property that distinguishes Lumos from its competitors and enables Lumos to charge premium pricing for its services. Without the protection of these proprietary assets, Lumos would be unable to maintain its market position.
C. Business Scale and Financial Position
6. As noted above, Lumos employs approximately 340 people as of the date of this declaration, including software engineers, machine-learning researchers, data scientists, product managers, sales professionals, and support staff. The majority of these employees are located at Lumos's San Jose headquarters, with additional personnel working from satellite offices and remote locations across the United States.
7. Lumos's 2022 annual recurring revenue ("ARR") was approximately $67 million, reflecting the significant value that institutional customers place on Lumos's proprietary analytical capabilities. Our revenue has grown substantially year-over-year since the company's early stages, driven by the differentiated performance of LumosEdge and the expansion of our customer base within the institutional finance sector.
8. To fund the research, development, and commercialization of its proprietary technology, Lumos has raised a total of approximately $185 million in venture capital funding through its Series D round. This capital was raised through successive funding rounds, each reflecting the growing value and promise of Lumos's technology and market opportunity. By way of specific example, Lumos's Series B round of $40 million was led by Ridgeline Ventures in 2019, at a time when LumosEdge was gaining significant market traction among early institutional customers. More recently, Lumos's Series D round of $72 million was led by Northstar Growth Equity and closed in June 2022, just months before the events giving rise to this litigation. The Series D round valued Lumos at a level reflecting the substantial maturity and competitive strength of the company's proprietary technology platform.
9. Lumos's investor base includes prominent venture capital and growth equity firms. Helen Trask, General Partner at Ridgeline Ventures, serves as a board observer for Lumos, and Jonathan Holt, Managing Director at Northstar Growth Equity, serves as a member of Lumos's Board of Directors. These investors committed substantial capital based in significant part on the value and defensibility of Lumos's proprietary technology. The significant capital raised by Lumos is a direct reflection of the substantial value of its proprietary technology assets and its competitive market position — assets that are now threatened by the conduct of Defendants described below.
II. LUMOS'S PROPRIETARY TECHNOLOGY AND TRADE SECRETS
A. Overview of Trade Secret Categories
10. Over the course of approximately six years, Lumos has developed, at great expense and through the dedicated efforts of its engineering and research teams, proprietary technology and confidential business information that constitutes its most valuable competitive assets. These assets are not publicly known or generally ascertainable, and Lumos has taken extensive and continuous measures to maintain their secrecy. Five categories of trade secrets are at issue in this matter, each of which I describe in detail below.
B. LumosEdge Ensemble Machine-Learning Model Architecture
11. The first category of trade secrets at issue is the LumosEdge ensemble machine-learning model architecture, including its proprietary neural-network topology and hyperparameter configurations. This architecture is the core intellectual engine of Lumos's entire product offering. It defines how LumosEdge combines multiple machine-learning models — including deep neural networks, gradient-boosted decision trees, and proprietary attention-based architectures — into an ensemble framework that produces predictive financial signals of exceptional accuracy and robustness.
12. The model architecture was developed through extensive research and development investment over a period of years, drawing on the specialized expertise of some of the most talented machine-learning engineers and researchers in the industry. The specific neural-network topologies, the manner in which individual model components are combined, and the hyperparameter configurations that govern the training and inference processes are the product of thousands of hours of experimentation, iteration, and refinement. This architecture is not publicly known or generally available in the field. Access to the model architecture specifications is strictly limited to authorized Lumos personnel who have a demonstrated need to know and who have executed appropriate confidentiality agreements.
C. Proprietary Feature-Engineering Libraries
13. The second category of trade secrets at issue is Lumos's proprietary feature-engineering libraries, which contain 2,847 individual engineered features used for financial signal generation. Feature engineering — the process of transforming raw data inputs into informative variables that enhance the predictive power of machine-learning models — is widely recognized as one of the most critical determinants of model performance in applied machine learning.
14. Lumos's feature-engineering libraries were developed iteratively over years of painstaking research and experimentation. Each of the 2,847 features represents a carefully designed transformation of one or more data inputs that has been empirically validated to contribute meaningfully to the predictive accuracy of LumosEdge's models. Many of these features encode proprietary insights about the relationships between alternative data sources and financial market behavior — insights that were arrived at only after extensive trial and error and that are not obvious or derivable from publicly available information. The library of engineered features represents a significant competitive moat. It cannot be independently replicated by a competitor without an investment of substantial time, capital, and expertise comparable to what Lumos itself expended to create it.
D. DataStream ETL Pipeline
15. The third category of trade secrets at issue is the "DataStream" ETL (extract, transform, load) pipeline codebase and associated data-schema definitions. DataStream is Lumos's proprietary system for ingesting, transforming, and loading data from a diverse array of sources into the LumosEdge analytical framework. DataStream covers ingestion from forty-three (43) separate alternative-data vendors, each of which provides data in different formats, with different delivery mechanisms, and subject to different quality and reliability characteristics.
16. The DataStream pipeline represents a proprietary system that enables Lumos to integrate disparate and complex data sources — including satellite imagery analytics, credit card transaction data, web traffic data, geolocation data, and many others — into a unified analytical framework suitable for consumption by LumosEdge's machine-learning models. The pipeline architecture, schema definitions, vendor integration logic, data-quality validation rules, and error-handling procedures are all confidential and proprietary to Lumos. Building a comparable ETL pipeline from scratch, including negotiating data access agreements with dozens of alternative-data vendors and developing the specialized integration logic for each, would require many months if not years of effort and millions of dollars in investment.
E. Curated Training Dataset
17. The fourth category of trade secrets at issue is a curated and labeled training dataset comprising approximately 14.6 terabytes of historical financial and alternative data. This dataset was assembled, cleaned, labeled, and curated at great expense and over a significant period of time. It includes historical market data, alternative data from dozens of vendors, proprietary labels reflecting actual financial outcomes, and metadata generated during the feature-engineering process.
18. The curated training dataset is foundational to the performance of LumosEdge's machine-learning models. The quality and comprehensiveness of training data is a critical determinant of model accuracy, and the extensive curation effort that Lumos invested in this dataset — including the removal of noise, the correction of errors, the imputation of missing values, and the careful alignment of disparate time series — is a core component of Lumos's competitive advantage. The dataset is stored on secured internal infrastructure with strict access controls, including role-based permissions, audit logging, and encryption at rest and in transit.
F. Confidential Product Roadmap
19. The fifth category of trade secrets at issue is Lumos's confidential product roadmap documents for the period 2023–2025, including, most critically, the unreleased specifications for "LumosEdge 3.0." The LumosEdge 3.0 roadmap document is thirty-seven (37) pages in length and details planned product features, architecture changes, integration plans, go-to-market strategies, and strategic initiatives that Lumos intends to pursue over the next several years.
20. The product roadmap constitutes some of the most competitively sensitive information in Lumos's possession. Disclosure of roadmap information to a direct competitor would cause severe and immediate strategic harm to Lumos by enabling the competitor to anticipate Lumos's product moves, preemptively develop competing features, and undercut Lumos's strategic positioning in the marketplace. The roadmap was distributed only to a limited number of senior Lumos employees and was marked "CONFIDENTIAL — INTERNAL USE ONLY."
G. Protective Measures
21. Lumos takes extensive and continuous measures to protect its trade secrets and confidential information from unauthorized access, use, and disclosure. These measures include, but are not limited to, the following:
(a) Restricting access to proprietary technology and confidential information on a strict need-to-know basis, with role-based access controls enforced through technical systems;
(b) Requiring all employees who have access to proprietary technology and confidential information to execute Lumos's standard Proprietary Information and Inventions Assignment Agreement ("PIIAA") as a condition of their employment;
(c) Employing comprehensive technical access controls and monitoring, including multi-factor authentication, encrypted storage, network segmentation, endpoint detection and response software, and audit logging of access to sensitive repositories and databases;
(d) Maintaining physical security protocols at Lumos's offices, including badge-access entry, visitor registration, and restricted access to server rooms and engineering workspaces; and
(e) Conducting regular security awareness training for all employees and periodic security audits of systems containing proprietary information.
22. These measures, taken together, demonstrate that Lumos treats its proprietary technology and confidential information as trade secrets and takes reasonable steps — indeed, extensive steps — to maintain their secrecy.
III. EMPLOYMENT AND DEPARTURE OF THE INDIVIDUAL DEFENDANTS
A. Rajiv Anand
23. Defendant Rajiv Anand, age 37, was employed by Lumos as Principal Engineer — ML Infrastructure from March 2019 through September 30, 2022. In this role, Anand was one of the key architects of the LumosEdge signal-generation engine. His responsibilities included designing, implementing, and optimizing the core machine-learning infrastructure that powers LumosEdge's ensemble models. He reported directly to Lumos's VP of Engineering, Rachel Fong, and worked closely with senior members of the machine-learning and data-engineering teams.
24. By virtue of his senior technical role, Anand had access to Lumos's most sensitive proprietary codebases and training data. Specifically, Anand had authorized access to the LumosEdge ensemble ML model architecture, including the proprietary neural-network topologies and hyperparameter configurations; the feature-engineering libraries; and the curated training datasets. His access to these materials was essential to the performance of his job duties, and he was trusted with this access precisely because of the confidentiality obligations he had undertaken.
25. Upon his hire in March 2019, Anand executed Lumos's standard PIIAA. The PIIAA includes the following restrictive covenant provisions, among others:
(a) An assignment of all inventions clause, pursuant to which Anand assigned to Lumos all right, title, and interest in any inventions, developments, or intellectual property created during or arising out of his employment at Lumos;
(b) A confidentiality and non-disclosure provision that survives the termination of Anand's employment for a period of three (3) years, during which Anand is prohibited from disclosing, using, or exploiting any Lumos confidential information or trade secrets;
(c) A non-solicitation of employees clause prohibiting Anand from soliciting, recruiting, or inducing any Lumos employee to leave Lumos's employ for a period of twelve (12) months following the date of Anand's separation from Lumos; and
(d) A covenant not to use or disclose Lumos trade secrets at any time, without limitation, in accordance with applicable trade secret law.
26. The PIIAA further requires that, upon the termination of employment for any reason, the departing employee must promptly return to Lumos all company materials, including but not limited to documents, files, data, code, hardware, and any other tangible or electronic materials containing or reflecting Lumos's proprietary information, and must certify in writing that all such materials have been returned and that no copies have been retained.
27. Anand resigned from Lumos on September 30, 2022, effective immediately, providing no advance notice. Critically, upon his departure, Anand did not return company materials and did not confirm the deletion of Lumos proprietary information from his personal devices, accounts, or storage media, as required by his PIIAA. Despite Lumos's subsequent requests, Anand has not complied with these obligations.
B. Serena Mallick
28. Defendant Serena Mallick, age 34, was employed by Lumos as Senior Software Engineer — Data Pipelines from June 2020 through October 14, 2022. In this role, Mallick was responsible for building, maintaining, and optimizing the proprietary DataStream ETL pipelines that ingest alternative financial data from dozens of third-party vendors into the LumosEdge platform. She was a key contributor to the DataStream system and possessed deep technical knowledge of its architecture, codebase, schema definitions, and vendor integration configurations.
29. By virtue of her role, Mallick had extensive access to the DataStream ETL pipeline codebase, its data-schema definitions, and the vendor integration configurations for multiple alternative-data providers. Her access to these materials was necessary for the performance of her job duties and was granted subject to the confidentiality and proprietary information protections described herein.
30. Upon her hire in June 2020, Mallick executed Lumos's standard PIIAA, which contains the same restrictive covenant provisions as those described above in connection with Anand's PIIAA, including the assignment of inventions clause, the confidentiality and non-disclosure provision surviving termination for three years, the non-solicitation of employees clause for twelve months post-separation, and the perpetual covenant not to use or disclose Lumos trade secrets.
31. Mallick resigned from Lumos on October 14, 2022, also without adequate advance notice. Like Anand, Mallick did not return company materials and did not confirm the deletion of Lumos proprietary information from her personal devices, accounts, or storage media upon her departure, in breach of her obligations under the PIIAA.
C. Derek Yun
32. Defendant Derek Yun, age 31, was employed by Lumos as Machine Learning Engineer II from January 2021 through October 21, 2022. In this role, Yun worked on model development, training, and evaluation within the LumosEdge platform. He had direct access to Lumos's proprietary training datasets, model hyperparameters, and feature-engineering libraries. Yun was involved in the day-to-day process of training and refining LumosEdge's machine-learning models and possessed detailed knowledge of the methodologies and data that underpin those models.
33. Upon his hire in January 2021, Yun executed Lumos's standard PIIAA, which contains the same restrictive covenant provisions as those executed by Anand and Mallick, as described above.
34. Yun resigned from Lumos on October 21, 2022, again without adequate advance notice. Like both Anand and Mallick, Yun did not return company materials and did not confirm the deletion of Lumos proprietary information from his personal devices, accounts, or storage media upon his departure, as required by his PIIAA.
D. Coordinated Nature of Departures
35. The departures of Anand, Mallick, and Yun occurred in rapid succession over a period of approximately three weeks — from September 30, 2022 through October 21, 2022. The timing and coordination of these departures, viewed in light of the events described in Sections IV and V below, strongly suggest a pre-planned and deliberate effort by the individual defendants to leave Lumos in a coordinated fashion and to establish a competing venture using Lumos's proprietary technology and trade secrets. I do not believe these near-simultaneous resignations were coincidental. In all my years leading Lumos, I have never experienced a coordinated departure of this nature, involving employees with such extensive access to the company's most sensitive proprietary materials.
IV. FORMATION OF NOVASPARK TECHNOLOGIES INC. AND COMPETITIVE THREAT
A. NovaSpark's Incorporation and Business
36. NovaSpark Technologies Inc. ("NovaSpark") was incorporated in the State of Delaware on October 3, 2022 — just three days after Anand's resignation from Lumos on September 30, 2022. I am informed and believe that the incorporation paperwork for NovaSpark was prepared and filed while Anand was still employed at Lumos or, at the very latest, within days of his departure, further evidencing that the formation of NovaSpark was pre-planned during the period of Anand's employment.
37. NovaSpark is headquartered at 88 Bluxome Street, Floor 4, San Francisco, California 94107. Defendant Rajiv Anand serves as the Chief Executive Officer and co-founder of NovaSpark.
38. Based on publicly available information, including NovaSpark's website and press releases, NovaSpark purports to offer AI-driven alternative-data analytics for hedge funds and proprietary trading desks. This is a market that directly and substantially overlaps with Lumos's core offering. Indeed, the description of NovaSpark's product and target customer base is remarkably similar to the description of LumosEdge and its target market, raising serious questions about the extent to which NovaSpark's product strategy and technology are derived from Lumos's proprietary materials.
B. Individual Defendants' Roles at NovaSpark
39. Defendant Serena Mallick joined NovaSpark as Vice President of Engineering on October 17, 2022 — just three days after her resignation from Lumos on October 14, 2022. The extraordinarily short interval between her departure from Lumos and her assumption of a senior leadership role at NovaSpark confirms that her departure from Lumos was pre-arranged and coordinated with Anand.
40. Defendant Derek Yun joined NovaSpark as a founding engineer on October 24, 2022 — just three days after his resignation from Lumos on October 21, 2022. As with Mallick, the near-instantaneous transition from Lumos to NovaSpark demonstrates that Yun's departure was part of a coordinated plan.
41. All three individual defendants — Anand, Mallick, and Yun — now hold senior positions at NovaSpark, a company that is a direct competitor to Lumos. Together, these three individuals possess deep and comprehensive knowledge of virtually every category of Lumos trade secret at issue in this case: the ML model architecture, the feature-engineering libraries, the DataStream ETL pipeline, the curated training datasets, and the confidential product roadmap. The concentration of this knowledge within a single competing startup is deeply concerning.
C. NovaSpark's Funding
42. On or about November 15, 2022, NovaSpark closed a $12 million seed funding round led by Blackpine Capital Partners LP. Franklin Osei is the Managing Partner of Blackpine Capital Partners LP, which is headquartered at 101 Mission Street, 38th Floor, San Francisco, California 94105. The closing of a $12 million seed round occurred approximately six weeks after NovaSpark's incorporation, at a time when the company was barely a month old and could not have developed significant proprietary technology of its own from scratch in such a brief period.
43. The rapid securing of significant venture capital funding for a startup of such recent vintage raises serious concerns about what assets, capabilities, or proprietary knowledge NovaSpark represented to its investors in order to secure such funding. A reasonable inference is that NovaSpark's pitch to Blackpine Capital Partners included, implicitly or explicitly, representations about the founders' deep expertise and access to technology and data that was in fact Lumos's proprietary information.
D. Direct Competitive Overlap
44. NovaSpark's stated business — AI-driven alternative-data analytics for institutional finance clients — directly competes with Lumos's core LumosEdge offering. Both companies target the same customer base: institutional asset managers, hedge funds, and proprietary trading desks that seek AI-powered analytics to gain a competitive edge in financial markets.
45. Given the departing employees' deep and comprehensive knowledge of Lumos's proprietary technology, NovaSpark's ability to rapidly build a competing product is inextricably linked to the trade secrets the individual defendants had access to during their tenure at Lumos. It strains credulity to believe that Anand, Mallick, and Yun could develop, in a matter of weeks or months, a product comparable to LumosEdge without drawing upon the proprietary knowledge, methodologies, and materials they acquired at Lumos over years of employment.
V. DISCOVERY OF DATA EXFILTRATION
A. IT Security Investigation
46. On December 12, 2022, Lumos's IT security team, in the course of a routine security audit, identified anomalous data-exfiltration activity associated with the corporate accounts and workstations of each of the three individual defendants during the final weeks of their respective employment at Lumos. Upon detecting these anomalies, the IT security team immediately escalated the matter for further investigation.
47. The investigation was conducted in coordination with Rachel Fong, Lumos's VP of Engineering, who possesses the technical expertise to understand the significance and sensitivity of the materials that were exfiltrated. Ms. Fong can attest to the technical details of the investigation and the nature of the compromised materials. The findings of this investigation are summarized below.
B. Specific Exfiltration Events
48. The investigation revealed the following specific acts of unauthorized data exfiltration by each of the individual defendants:
Rajiv Anand:
49. On September 26, 2022 — four days before his resignation from Lumos — Anand transferred approximately 3.2 gigabytes (GB) of files from Lumos's internal code repositories and configuration management systems to a personal Hollowell Health IT account. Lumos's access logs confirm that the transferred files included model architecture specifications, hyperparameter configuration files, and related technical documentation associated with the LumosEdge ensemble ML model. These files contain some of the most sensitive and valuable proprietary information in Lumos's possession, representing the core intellectual property underlying LumosEdge's predictive capabilities.
50. On September 28, 2022 — two days before his resignation from Lumos — Anand accessed and forwarded the confidential thirty-seven (37) page "LumosEdge 3.0 Product Roadmap" document, in PDF format, to his personal email address, r.anand.personal@vaultmail.com. Lumos's email monitoring systems recorded this transmission. The LumosEdge 3.0 Product Roadmap is among the most competitively sensitive documents in Lumos's possession, containing detailed plans for future product features, architectural enhancements, and strategic initiatives covering the period from 2023 through 2025.
Serena Mallick:
51. On October 10, 2022 — four days before her resignation from Lumos — Mallick cloned the entire DataStream ETL pipeline repository from Lumos's internal version-control system to a personal external USB drive. Lumos's endpoint-detection software recorded the connection of the USB drive (identified by serial number WD-4837291) and the transfer of approximately 780 megabytes (MB) of data to the device. The transferred repository contains the complete DataStream ETL pipeline codebase, including source code, data-schema definitions, vendor integration configurations, and related documentation. This codebase represents the entirety of Lumos's proprietary data-ingestion infrastructure.
Derek Yun:
52. On October 18, 2022 — three days before his resignation from Lumos — Yun downloaded forty-seven (47) proprietary Jupyter notebooks from Lumos's internal development environment to a personal Larkspur Technologies account. Lumos's data-loss-prevention systems detected and logged this transfer. The Jupyter notebooks contain feature-engineering code, model-training scripts, experimental results, and associated commentary documenting the rationale behind specific modeling choices. These notebooks encapsulate significant proprietary knowledge about Lumos's feature-engineering methodologies and model-training processes.
C. Failure to Return Materials
53. As noted above, none of the three individual defendants returned company materials or confirmed the deletion of Lumos proprietary information from their personal devices, accounts, or storage media upon their departure from Lumos, as required by the terms of their respective PIIAAs. Despite Lumos's subsequent demands, none of the individual defendants has complied with these obligations as of the date of this declaration.
54. Upon discovering the data-exfiltration events described above, Lumos acted promptly to assess the scope of the breach and to seek legal counsel. On December 15, 2022, Lumos retained the law firm of Calderwood & Harkness LLP to investigate the matter and to advise on potential litigation. Calderwood & Harkness has been engaged in an intensive investigation since that date, the results of which inform this declaration and the accompanying application for emergency relief.
VI. IRREPARABLE HARM TO LUMOS
A. Nature of Harm
55. The continued possession, use, and potential disclosure of Lumos's trade secrets by Defendants causes and threatens to cause irreparable harm to Lumos that cannot be adequately compensated by monetary damages alone. It is well established that trade secrets, once disclosed, lose their status as trade secrets and cannot be "unlearned" or recalled. The unauthorized possession of Lumos's trade secrets by the individual defendants and their use by or on behalf of NovaSpark creates a continuing and escalating risk that these trade secrets will be further disseminated, incorporated into competing products, or otherwise exploited in ways that permanently destroy their value as proprietary assets.
56. The harm that Lumos faces is not speculative. Defendants have already taken Lumos's trade secrets. Defendants are already building a directly competing product. And Defendants already have the financial resources to bring that product to market. The risk of irreparable harm is immediate, concrete, and severe.
B. Competitive Harm
57. NovaSpark is in direct competition with Lumos for the same institutional customers — asset managers, hedge funds, and proprietary trading desks — that constitute Lumos's core revenue base. If NovaSpark uses Lumos's proprietary technology to develop and launch a competing product, Lumos will lose the competitive advantage that it has built over six years and at a cost of hundreds of millions of dollars in investment. Customers who are offered a product built on technology substantially identical to LumosEdge's will have reduced incentive to contract with Lumos, and Lumos will face the impossible task of competing against its own proprietary technology.
58. Moreover, the harm from the loss of trade-secret protection is inherently difficult, if not impossible, to quantify. Once a trade secret is incorporated into a competitor's product, it becomes exceedingly difficult to disentangle the competitor's independently developed technology from the misappropriated trade secrets. Lost customer relationships, diminished pricing power, and eroded market position are harms that defy precise monetary calculation and are therefore irreparable.
C. Strategic and Reputational Harm
59. The disclosure of the LumosEdge 3.0 product roadmap to a direct competitor would undermine Lumos's multi-year strategic product planning. The roadmap reflects Lumos's assessment of market opportunities, its planned technology investments, and its competitive positioning strategy for the period 2023 through 2025. If NovaSpark possesses this roadmap — which Anand forwarded to his personal email account just days before his departure — NovaSpark will be able to anticipate and preempt Lumos's product moves, neutralizing the strategic advantages that Lumos's planned innovations were designed to create.
60. In addition, Lumos's investors — including Ridgeline Ventures and Northstar Growth Equity — have invested a total of $185 million in Lumos based in significant part on the value and defensibility of Lumos's proprietary technology. If Lumos's trade secrets are dissipated through misappropriation, the foundation of that investment will be undermined. Loss of investor confidence could impair Lumos's ability to raise additional capital, attract and retain top talent, and pursue its strategic objectives. Similarly, the confidence of Lumos's institutional customers — who entrust Lumos with their most sensitive portfolio data and rely on LumosEdge for mission-critical investment decisions — could be severely damaged if those customers learn that Lumos's proprietary technology has been compromised. The resulting reputational harm would further compound the irreparable injury to Lumos.
D. Urgency.
61. Every day that Defendants retain and potentially use Lumos's trade secrets compounds the harm to Lumos. NovaSpark is actively building a competing product, and there is no indication that Defendants have ceased using or relying upon Lumos's proprietary materials. NovaSpark has the financial resources — a $12 million seed round — to accelerate its product development, hire additional engineers, and bring a competing product to market in a compressed timeframe.
62. Absent immediate judicial intervention, the damage to Lumos will be irremediable. Each day that passes without the protection of a temporary restraining order increases the risk that Lumos's trade secrets will be further disseminated, permanently incorporated into NovaSpark's products, or disclosed to additional third parties, including NovaSpark's investors, prospective customers, or new hires. Once that occurs, no amount of monetary damages can restore the status quo ante.
VII. NEED FOR EMERGENCY RELIEF
63. Given the nature and sensitivity of the proprietary information taken by Defendants, the direct and immediate competitive threat posed by NovaSpark, the documented evidence of data exfiltration, the individual defendants' refusal to return company materials, and the ongoing and accelerating risk of irreparable harm, Lumos respectfully submits that emergency relief in the form of a temporary restraining order is both necessary and fully justified.
64. Lumos respectfully requests that the Court order Defendants to:
(a) Immediately cease all use, disclosure, dissemination, and exploitation of Lumos's proprietary information and trade secrets, including but not limited to the LumosEdge ensemble ML model architecture, the proprietary feature-engineering libraries, the DataStream ETL pipeline codebase and data-schema definitions, the curated training dataset, and the confidential product roadmap documents;
(b) Immediately preserve all electronic devices, storage media, cloud accounts, email accounts, and any other repositories containing or potentially containing Lumos materials, and refrain from destroying, deleting, altering, moving, or concealing any documents, files, data, code, or other materials related to Lumos's proprietary technology; and
(c) Refrain from soliciting, recruiting, or inducing any current Lumos employee to leave Lumos's employ or to join NovaSpark or any other enterprise affiliated with any Defendant, in accordance with the non-solicitation provisions of the individual defendants' respective PIIAAs.
65. Without such relief, Lumos faces the imminent and concrete prospect of losing trade secrets that are the very foundation of its business and that represent years of intensive development and hundreds of millions of dollars in investment by Lumos and its investors. The balance of hardships overwhelmingly favors the issuance of a temporary restraining order: while Defendants may be temporarily inconvenienced by an order requiring them to refrain from using another company's stolen property, Lumos faces the permanent destruction of its most valuable assets.
66. Lumos is prepared to post a bond in an amount the Court deems appropriate as security for any damages that Defendants may sustain by reason of the temporary restraining order, should the Court ultimately determine that the order was wrongfully issued. Lumos submits, however, that any bond should be modest given the overwhelming evidence of misappropriation and the minimal hardship to Defendants of an order that merely requires them to refrain from using property that is not theirs.
I declare under penalty of perjury under the laws of the State of California that the foregoing is true and correct.
Executed on January 19, 2023, at San Jose, California.
________________________________________
Priya Chandrasekaran Chief Executive Officer Lumos Analytics Inc. 4700 Orchard Parkway, Suite 300 San Jose, California 95134
Prepared by:
CALDERWOOD & HARKNESS LLP
Marcus Delgado, Responsible Partner Catherine Bloom, Billing Partner Wendy Schultz, Lead Associate Omar Siddiqui, Working Associate
555 California Street, 40th Floor San Francisco, California 94104
Attorneys for Plaintiff Lumos Analytics Inc.
C&H Matter Number: 1008-00005
LUMOS ANALYTICS INC., a Delaware corporation, | Case No.: 23-CV-004871
Plaintiff, | DECLARATION OF PRIYA CHANDRASEKARAN IN SUPPORT OF PLAINTIFF'S EX PARTE APPLICATION FOR TEMPORARY RESTRAINING ORDER AND ORDER TO SHOW CAUSE RE: PRELIMINARY INJUNCTION
v.
NOVASPARK TECHNOLOGIES INC., a Delaware corporation; RAJIV ANAND, an individual; SERENA MALLICK, an individual; and DEREK YUN, an individual,
Defendants.