DCT

3:26-cv-05930

Wwwailaw Corp v. Butler Labs Inc

Log In
Key Events
Amended Complaint
complaint Intelligence

I. Executive Summary and Procedural Information

  • Parties & Counsel:
  • Case Identification: 3:26-cv-05930, N.D. Cal., 09/21/2026
  • Venue Allegations: Venue is alleged to be proper based on Defendant having a regular and established principal place of business in the Northern District of California.
  • Core Dispute: Plaintiff alleges that Defendant’s AI-powered legal technology platform infringes a patent related to a method for transforming unstructured data into structured, long-form documents using a specific AI workflow.
  • Technical Context: The technology resides in the field of generative AI for document drafting, specifically concerning architectural improvements over standard Retrieval-Augmented Generation (RAG) systems to enhance output quality, length, and performance.
  • Key Procedural History: Plaintiff alleges it provided Defendant with notice of the patent-in-suit and its alleged infringement via correspondence on May 18, 2026, and June 1, 2026. The complaint also notes that during patent prosecution, the applicants overcame a patent eligibility rejection under 35 U.S.C. §101 by tying the invention to a specific physical device, which may be relevant to future validity arguments.

Case Timeline

Date Event
2024-10-25 '932 Patent Priority Date
2025-07-21 '932 Patent Prosecution: Response to Office Action referenced in complaint
2025-11-04 '932 Patent Issue Date
2026-03-05 Plaintiff allegedly marks its website with the patent number
2026-05-18 Plaintiff sends initial notice of infringement to Defendant
2026-06-01 Plaintiff sends follow-up correspondence to Defendant
2026-09-21 Complaint Filing Date

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

U.S. Patent No. 12,461,932 - "Method and System for Transforming Data Using Artificial Intelligence to Generate Content"

  • Patent Identification: U.S. Patent No. 12,461,932, "Method and System for Transforming Data Using Artificial Intelligence to Generate Content," issued November 4, 2025 (the "'932 Patent").

The Invention Explained

  • Problem Addressed: The patent's background section describes conventional AI content generation for tasks like drafting legal documents as being potentially "inaccurate, time-consuming, inefficient, and resource-wasteful" ’932 Patent, col. 1:17-26 The specification further identifies specific technical problems with Retrieval-Augmented Generation (RAG) systems, noting they can "miss items of interest and be less comprehensive" ’932 Patent, col. 22:55-57 Compl. ¶51 Another identified problem is the computational resource consumption required to transform unstructured data for processing by an AI engine ’932 Patent, col. 4:1-5
  • The Patented Solution: The invention claims a specific, multi-step workflow to improve upon these limitations. Instead of feeding raw unstructured text to a large language model (LLM), the method first transforms the "unstructured input data into structured data" ’932 Patent, claim 1(b) To address processing delays, it claims to execute "a plurality of functions in parallel to reduce execution time" ’932 Patent, claim 1(d) Crucially, it introduces a "context-aware feedback loop" that detects the reason why an LLM has stopped generating content and then conditionally continues the process, allowing for the creation of long, cohesive documents that might otherwise be limited by LLM output constraints ’932 Patent, claim 1(e) ’932 Patent, col. 19:1-26
  • Technical Importance: The claimed method purports to offer a more robust and efficient architecture than standard RAG, enabling the generation of complex, long-form documents with higher accuracy and consistency, particularly in specialized domains like law Compl. ¶53 Compl. ¶64

Key Claims at a Glance

  • The complaint asserts independent claims 1 (method), 19 (computer-readable medium), and 20 (system) Compl. ¶77 Compl. ¶193 Compl. ¶197
  • Independent Claim 1, the core method claim, includes the following essential elements:
    • (a) receiving unstructured input data;
    • (b) transforming the unstructured input data into structured data using an LLM and a parsing algorithm;
    • (c) generating, based on the structured data, one or more outputs to be processed by an AI engine;
    • (d) generating a document by converting the outputs, which includes executing a plurality of functions in parallel to reduce execution time;
    • (e) in response to detecting generation has stopped, executing a context-aware feedback loop using a prompt that detects a reason for the stoppage;
    • (f) in response to detecting generation stopped because it was complete, providing the final document for presentation; and
    • (g) in response to detecting generation stopped due to an output constraint, re-executing the prompt with a partially-generated document string.
  • The complaint reserves the right to assert infringement of numerous dependent claims and other claims based on discovery Compl. ¶264

III. The Accused Instrumentality

Product Identification

  • The "Eve Legal Accused System," which is Defendant's "Eve" AI platform, accessible at www.eve.legal Compl. ¶66 Compl. ¶67

Functionality and Market Context

  • The Eve Legal Accused System is an AI-powered platform designed for legal professionals to automate tasks such as document drafting (complaints, motions, discovery responses), case analysis, and medical chronology generation (Compl. ¶¶7, 66).
  • The complaint alleges the system uses various LLMs, including Anthropic's Claude models, as computational resources Compl. ¶68 A central allegation is that Defendant has transitioned the system from a standard RAG architecture to a more complex "agentic architecture" that operates an "orchestration layer" to manage multi-stage and multi-turn model operations, which allegedly practices the patented method (Compl. ¶70; Compl. ¶71). The complaint references an online case study stating that Defendant "built its own infrastructure and custom harness around Claude" Compl. ¶71 Compl. ¶136 The complaint also provides a screenshot from Defendant's website describing its capability to "Draft complaints, motions, and more" using a user's templates and case context Compl. ¶83

IV. Analysis of Infringement Allegations

'932 Patent Infringement Allegations

Claim Element (from Independent Claim 1) Alleged Infringing Functionality Complaint Citation Patent Citation
(a) receiving unstructured input data; The system receives uploaded user documents, such as case files, transcripts, evidence, and notes, which constitute unstructured input data. ¶¶81-84; ¶92 col. 4:18-20
(b) transforming the unstructured input data into structured data, wherein the transforming is performed using a large language model and a parsing algorithm... The system's orchestration layer allegedly uses LLMs (like Claude Sonnet) and a "parsing algorithm" to extract information from uploaded documents and organize it into structured formats, such as JSON, for downstream use. ¶93; ¶97; ¶105 col. 4:15-18
(c) generating, based on the structured data, one or more outputs configured to be processed by an artificial intelligence engine; The system allegedly generates outputs from the structured data, such as extracted facts, claims, and other case details, that are configured as inputs for subsequent AI processing and document generation. ¶118; ¶121 col. 4:32-35
(d) generating, using the artificial intelligence engine, a document... wherein generating the document includes executing a plurality of functions in parallel to reduce execution time... The system allegedly generates documents like discovery responses or medical chronologies by processing multiple individual requests or records concurrently, rather than sequentially, to reduce overall processing time. ¶127; ¶128; ¶131 col. 6:51-56
(e) in response to detecting that generation of the document stopped, executing a context-aware feedback loop using a prompt that detects a reason why the generation of the document stopped; The system's custom harness allegedly evaluates the stop_reason field returned by the Claude API to detect why generation stopped, distinguishing between natural completion and hitting an output constraint. ¶138; ¶145; ¶146 col. 19:40-50
(f) in response to detecting that the generation of the document stopped due to the generation... being complete, providing the document to a computing device for presentation...; When the stop_reason indicates natural completion (e.g., 'end_turn'), the system allegedly determines the workflow is complete and provides the final document to the user's web interface for presentation. ¶151; ¶153; ¶155 col. 6:57-61
(g) in response to detecting that the generation of the document stopped due to an output constraint of the large language model, re-executing the prompt with a partially-generated document string. When the stop_reason indicates an output constraint was met (e.g., 'max_tokens'), the system allegedly retains the partially generated content and submits a new request to the LLM to continue generation from where it left off. ¶158; ¶162; ¶168 col. 19:51-60

Identified Points of Contention

  • Technical Questions: A central technical question is whether the Accused System's "agentic architecture" actually performs the specific "context-aware feedback loop" as claimed. The complaint's allegations heavily rely on inferences drawn from publicly available technical documentation for the Claude API, particularly the stop_reason field Compl. ¶140 A screenshot from Anthropic's documentation is included in the complaint to support this theory Compl. ¶140 The court will likely need to resolve whether Defendant’s system uses this API feature in the manner required by claims 1(e), 1(f), and 1(g).
  • Scope Questions: The case may turn on the definition of "parallel" execution. The complaint alleges parallel processing on "information and belief," reasoning that it is necessary to achieve the performance required for large-scale document generation Compl. ¶130 The question for the court will be whether the Accused System's architecture involves true concurrent execution of generation functions, as opposed to merely asynchronous or sequential processing, and whether the latter falls within the scope of the claim.
  • Scope Questions: The construction of what constitutes "transforming... into structured data" will be another point of focus. The complaint alleges this is met by using an LLM to extract information and place it into formats like JSON Compl. ¶105 Compl. ¶107 The question is whether this process meets the claim limitation of using both a "large language model and a parsing algorithm," and what level of organization is required to be considered "structured."

V. Key Claim Terms for Construction

"context-aware feedback loop"

  • Context and Importance: This term appears to be the core of the claimed invention, distinguishing it from simpler methods of handling LLM output. Infringement of elements 1(e), 1(f), and 1(g) depends entirely on whether the Accused System implements a process that meets this definition, specifically one that "detects a reason" for a generation stoppage and branches its logic accordingly. Practitioners may focus on this term because the complaint's theory of infringement is heavily based on inferring this functionality from third-party API documentation.
  • Intrinsic Evidence for a Broader Interpretation: The specification introduces the concept as a "novel context aware feedback loop for the generation of content" Compl. ¶42, which could support an argument that any loop aware of generation state is covered.
  • Intrinsic Evidence for a Narrower Interpretation: The claim language itself is highly specific, requiring a loop that "detects a reason" and then executes different steps based on that reason (completion vs. output constraint). The detailed description of the feedback loop system could be used to argue for a narrow construction tied to this specific conditional logic ’932 Patent, col. 19:1-60

"executing a plurality of functions in parallel"

  • Context and Importance: This term defines a specific performance-oriented architectural requirement. If the Accused System is found to operate sequentially, it would not infringe this element of the claim. The complaint alleges this on information and belief based on performance needs, making it a likely area of factual dispute Compl. ¶130
  • Intrinsic Evidence for a Broader Interpretation: The stated purpose is to "reduce execution time" ’932 Patent, claim 1(d) This could support an interpretation where any form of overlapping execution that speeds up the process, such as pipelining or asynchronous calls, constitutes "parallel" execution.
  • Intrinsic Evidence for a Narrower Interpretation: The patent describes addressing a "processing delay that can result when multiple AI document-generation functions are performed sequentially" Compl. ¶61 This could support an argument that the term requires true, concurrent processing of independent functions, not just an asynchronous workflow.

VI. Other Allegations

  • Indirect Infringement: The complaint alleges induced infringement under 35 U.S.C. § 271(b), stating that Defendant instructs and encourages its customers to use the Accused System in an infringing manner through tutorials, workflows, videos, and help pages Compl. ¶191 Compl. ¶273 The complaint identifies specific customers as alleged direct infringers Compl. ¶274
  • Willful Infringement: The complaint alleges willful infringement based on Defendant’s alleged continued infringement after receiving notice of the '932 Patent on May 18, 2026, and June 1, 2026 Compl. ¶2 Compl. ¶267 Compl. ¶277

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

  • A central evidentiary question will be one of functional implementation: does the Accused System’s "custom harness" for the Claude LLM actually use the API's stop_reason field to implement the specific, reason-detecting, and conditional logic of the "context-aware feedback loop" as required by claim 1? The complaint builds a detailed infringement theory based on public API documentation, but proof will likely depend on discovery of Defendant's nonpublic source code and internal operations.
  • A key issue will be one of definitional scope: does the term "transforming... into structured data" encompass the alleged process of using an LLM to extract information and organize it into a format like JSON, or does the claim's requirement of both an "LLM and a parsing algorithm" imply a more distinct, two-part process that the Accused System may not perform?
  • A final question will be one of architectural fact: does the Accused System actually perform "parallel" execution of document generation functions, or does its architecture rely on a sequential or asynchronous process that falls outside the scope of that claim limitation? The complaint infers this feature from performance necessities, which will require technical evidence from discovery to substantiate.