Poulin Holdings LLC v. Databricks Inc

I. Executive Summary and Procedural Information

• Parties & Counsel:
  • Plaintiff: Poulin Holdings, LLC (New Hampshire)
  • Defendant: Databricks, Inc. (Delaware)
  • Plaintiff's Counsel: Hecht Partners LLP
• Case Identification: 4:26-cv-00245, E.D. Tex., 03/10/2026
• Venue Allegations: Plaintiff alleges venue is proper in the Eastern District of Texas because Defendant has committed acts of infringement in the district and maintains a regular and established place of business there, specifically a corporate office in Plano, Texas.
• Core Dispute: Plaintiff alleges that Defendant's Databricks Lakehouse Platform infringes two patents related to systems and methods for building, storing, managing, and governing access to predictive computer models.
• Technical Context: The dispute centers on the architecture of machine learning operations (MLOps) platforms, which manage the lifecycle of predictive models from data ingestion to deployment and governance in enterprise environments.
• Key Procedural History: The complaint does not mention any prior litigation, Inter Partes Review (IPR) proceedings, or specific licensing history concerning the patents-in-suit.

Case Timeline

II. Technology and Patent(s)-in-Suit Analysis

U.S. Patent No. 8,160,977 - "Collaborative predictive model building"

The Invention Explained

• Problem Addressed: The patent describes a state of the art where predictive modeling involved fragmented, ad-hoc workflows, making it difficult to process large data streams into stable, reusable training datasets Compl. ¶29 Models were often "black box" processes, lacking repeatability and objective quality comparison Compl. ¶29
• The Patented Solution: The invention proposes a specific, non-conventional computer architecture for managing the predictive model lifecycle Compl. ¶30 The core concept involves using two distinct databases: a "first database" to store processed "precursor data" (i.e., features for training) and a "second database" to store the trained predictive models themselves as searchable, auditable records '977 Patent, abstract '977 Patent, col. 1:25-39 This separation is intended to make model training repeatable and to treat models as discoverable assets rather than static files Compl. ¶33 The system also calculates and stores accuracy metrics to allow for objective model comparison Compl. ¶30
• Technical Importance: This architectural approach aims to improve a computer's ability to manage the entire model lifecycle, from data preparation to performance validation and reuse, addressing governance and auditability challenges in enterprise-scale machine learning Compl. ¶30

Key Claims at a Glance

• The complaint asserts at least independent Claim 1 Compl. ¶55
• Essential elements of Claim 1 include:
  • decomposing by one or more computer systems a data input stream to build a first database of precursor data to build at least one predictive model;
  • building by the one or more computer systems at least one model generated by a model building process using the precursor data in the precursor database, with the at least one model being a model that produces predictions of the likelihood of an event occurring in the future;
  • storing by the one or more computer systems the at least one model in a second database that stores models, with the second database being searchable to permit the at least one model in the second database to be accessed by users; and
  • calculating by the one or more computer systems accuracy of the at least one model against historical data.
• The complaint does not explicitly reserve the right to assert dependent claims but alleges infringement of "one or more claims" Compl. ¶55

U.S. Patent No. 9,009,090 - "Predictive model database with predictive model user access control rights and permission levels"

The Invention Explained

• Problem Addressed: The patent addresses a "governance bottleneck" in enterprise environments where a proliferation of predictive models lacked a secure, rights-conditioned method for discovery and use Compl. ¶38 Existing systems were described as largely binary, where access was either universally open or entirely closed Compl. ¶38
• The Patented Solution: The invention claims a technical improvement by integrating a rights-conditioned access workflow directly into the model database record '090 Patent, abstract Compl. ¶39 The system stores models with associated permission levels that specify access rights Compl. ¶40 When a user selects a model from search results, the system synthesizes the user's identity and the model's access rights to dynamically generate a message with enabling options (e.g., "request access" or immediate use) consistent with the user's permissions Compl. ¶¶39-40 This is positioned as a specialized improvement to computer security that forces the system to act as a "gatekeeper" of its own data records '090 Patent, col. 1:40-45 Compl. ¶39
• Technical Importance: The solution provides a machine-governed way to manage access to predictive models, moving beyond simple information display to a control-flow operation that dynamically determines and enables user actions based on stored permissions Compl. ¶39

Key Claims at a Glance

• The complaint asserts at least independent Claim 1 Compl. ¶62
• Essential elements of Claim 1 include:
  • storing a plurality of models in a database... the models having permission levels that specify the access rights to control access to the models by users;
  • receiving a search query to search the database that stores the plurality of models...;
  • executing the search query to produce search results that are provided based on search terms included in the search query;
  • sending by a computer system the search results to a user system;
  • receiving by the computer system a selection of a model from the search results; and
  • sending by a computer system a message to the user system, the message based on the access rights associated with the model, with the message including at least one option to access the model.
• The complaint does not explicitly reserve the right to assert dependent claims but alleges infringement of "one or more claims" Compl. ¶62

III. The Accused Instrumentality

Product Identification

• The accused instrumentality is the "Databricks Lakehouse Platform," which includes, without limitation, Databricks Feature Store, MLflow tracking and evaluation, MLflow Model Registry and/or Models in Unity Catalog, Unity Catalog governance and permissions, and Databricks Marketplace Compl. ¶¶52-54

Functionality and Market Context

• The complaint alleges the Databricks platform is an integrated system for data engineering and machine learning Compl. ¶44 The "Feature Store" is alleged to transform input data into persisted "feature tables" that function as precursor datasets for model training Compl. ¶45 "MLflow Model Registry" and "Models in Unity Catalog" are alleged to store trained models as first-class objects in a database dedicated to models, which is searchable via interfaces like "Catalog Explorer" Compl. ¶48 The platform also allegedly provides for model evaluation against historical data to calculate and log accuracy metrics Compl. ¶47
• "Unity Catalog" is alleged to provide centralized governance, associating models with access control policies and permission levels Compl. ¶49 "Databricks Marketplace" is alleged to provide a searchable marketplace where users can select assets and receive a "rights-conditioned response" that presents options for access (e.g., request or subscription workflows) based on the user's permissions Compl. ¶50

IV. Analysis of Infringement Allegations

The complaint references claim charts that are to be provided as exhibits, but these exhibits were not filed with the complaint Compl. ¶61 Compl. ¶69 The following analysis is constructed from the narrative infringement allegations provided in the body of the complaint. No probative visual evidence provided in complaint.

• '977 Patent Infringement Allegations

• Identified Points of Contention:

  • Scope Questions: The analysis may raise the question of whether Databricks' "Feature Store," which stores "feature tables," meets the claim limitation of "a first database of precursor data." Similarly, a question may arise as to whether the "MLflow Model Registry" constitutes "a second database that stores models" as distinct from the first.
  • Technical Questions: A central question may be whether the alleged separation between feature storage and model storage in the Databricks platform is technically equivalent to the two-database architecture described in the patent.

• '090 Patent Infringement Allegations

• Identified Points of Contention:
  • Scope Questions: The dispute may center on whether the workflows in Databricks Marketplace, which present "request or subscription" options, meet the claim limitation of "sending... a message... including at least one option to access the model."
  • Technical Questions: A key question will be whether the alleged access control in Unity Catalog and Marketplace is technically a "message based on the access rights" that is dynamically generated post-selection, as described in the patent, or a more conventional pre-configured access gate.

V. Key Claim Terms for Construction

• The Term: "a first database of precursor data" '977 Patent, Claim 1

  • Context and Importance: This term is foundational to the '977 Patent's two-database architecture. The definition will be critical to determining if Databricks' "Feature Store" Compl. ¶56, which stores "feature tables," infringes. Practitioners may focus on whether a collection of "feature tables" constitutes a "database" and whether those features are "precursor data" as contemplated by the patent.
  • Intrinsic Evidence for Interpretation:
    • Evidence for a Broader Interpretation: The specification describes decomposing a "data input stream" to "build a database of precursor data" '977 Patent, abstract This could be argued to cover any structured, persistent collection of processed data intended for model training, such as a feature store.
    • Evidence for a Narrower Interpretation: The detailed description discusses collecting data, normalizing it, and formatting it into a "combined data file" (e.g., wordFreqs.dat) for use by the model building process '977 Patent, col. 4:50-58 This could support a narrower definition tied to a more specific file-based or monolithic database structure, potentially distinct from a distributed feature table system.

• The Term: "a message based on the access rights associated with the model" '090 Patent, Claim 1

  • Context and Importance: This term is the core of the '090 Patent's claimed technical improvement over simple access control. The infringement analysis for Databricks Marketplace hinges on whether its "rights-conditioned response" Compl. ¶50 is such a "message." Practitioners may focus on whether the claim requires the message itself to be dynamically constructed based on a real-time synthesis of user and model rights, or if it can cover a pre-determined response based on a user's role.
  • Intrinsic Evidence for Interpretation:
    • Evidence for a Broader Interpretation: The claim requires the message to be "based on the access rights," which could arguably cover any system response that is conditioned by permissions, such as presenting a 'subscribe' button to one user and an 'access' button to another.
    • Evidence for a Narrower Interpretation: The complaint argues the invention "executes a control-flow operation where the message returned to the user is dynamically determined" Compl. ¶39 The specification discusses generating options like "'Make Bid' or 'Pay the ASK' price" '090 Patent, col. 10:40-43 Fig. 4B This may support a narrower construction requiring a dynamic, transaction-oriented message generation process, rather than simply displaying different static options based on permissions.

VI. Other Allegations

• Indirect Infringement: The complaint alleges inducement of infringement for both patents, asserting that Defendant provides instructions, documentation, training materials, marketing, and technical support that encourage and instruct customers and end users to use the accused platform in an infringing manner Compl. ¶76 Compl. ¶86 It also alleges contributory infringement, stating the accused components are not staple articles of commerce and are especially made or adapted for use in an infringing manner Compl. ¶77 Compl. ¶87
• Willful Infringement: The complaint alleges willful infringement for both patents based on Defendant's knowledge of the patents "at minimum... as of the filing and service of this Complaint" Compl. ¶78 Compl. ¶88 This frames the willfulness allegation as being based on post-suit conduct.

VII. Analyst's Conclusion: Key Questions for the Case

• A core issue will be one of architectural equivalence: Does the Databricks platform, with its "Feature Store" for precursor data and "Model Registry" for models, implement the specific two-database architecture required by the '977 Patent, or does it represent a technically distinct system for managing machine learning assets?
• A second central issue will be one of functional scope: Does the "rights-conditioned response" in Databricks Marketplace perform the specific, dynamic, post-selection message generation claimed by the '090 Patent, or is it a conventional access control mechanism that falls outside the claimed "control-flow operation"?
• An evidentiary question will be one of causation and control: To the extent infringement relies on the actions of Databricks' customers, the case may turn on what evidence shows Defendant's platform directly performs the claimed method steps versus merely enabling customers to do so, which will be relevant to the direct and indirect infringement claims.