PTAB

PGR2025-00075

OneSource Solutions International, Inc. v. Hippocratic AI, Inc.

Title: Multi-Turn Conversational System with Second Opinion Module
Brief Description: The ’371 patent discloses a system using a first large language model (LLM) to conduct a multi-turn, checklist-guided conversation with a user, while a second LLM, the “second opinion module,” performs research in parallel to validate information. Control logic modules analyze the conversation to detect triggers, insert questions, and escalate to human experts.

Prior Art Relied Upon: Brown (Patent 9,824,188), the ’968 application (Application # 2024/0185968), and the ’651 application (Application # 2023/0245651).
Core Argument for this Ground:
- Prior Art Mapping: Petitioner argued that the primary combination teaches all limitations of the independent claims. Brown disclosed a medical chatbot capable of multi-turn dialogue and user intent determination. The ’968 application taught a checklist-based query system for performing a predefined dialogue. The ’651 application taught a parallel-running analysis module configured to collaborate with a conversational AI to perform in-depth analysis and generate recommendations, functioning as a second opinion.
- Motivation to Combine: A person of ordinary skill in the art (POSA) would combine the medical chatbot of Brown with the checklist enforcement of the ’968 application and the parallel analysis module of the ’651 application. The motivation was the predictable goal of creating a more robust, reliable, and safe conversational system that ensures required tasks are completed while independently verifying information.
- Expectation of Success: A POSA would have had a reasonable expectation of success, as combining these known elements would predictably result in a system comprising a checklist-guided interface, a control logic module, and a parallel second-opinion mechanism, as claimed.

Prior Art Relied Upon: The ’644 patent (Patent 10,748,644), the ’694 patent (Patent 11,348,694), and the ’854 patent (Patent 11,977,854).
Core Argument for this Ground:
- Prior Art Mapping: Petitioner asserted this alternative combination also rendered the independent claims obvious. The ’644 patent taught a mental health assessment system with a conversation interface, control logic, and a “runtime model server logic” that concurrently evaluates patient signals, mapping to the claimed second opinion module. The ’694 patent taught a checklist-based conversational system for determining a medical condition. The ’854 patent taught using an LLM trained with at least one thousand gradient update iterations.
- Motivation to Combine: A POSA would have been motivated to combine the comprehensive system of the ’644 patent with the checklist-based approach of the ’694 patent to ensure structured data collection. Incorporating the standard LLM training taught by the ’854 patent was a basic step to implement the system.

Core Argument for this Ground: Petitioner argued the ’371 patent fails to teach a POSA how to make and use the full scope of the invention without undue experimentation. The specification allegedly describes critical components as functional "black boxes." For example, it does not disclose the algorithms, data schemas, or synchronization protocols for the “second-opinion module” to perform research in parallel, how “control signals” are generated and integrated into the conversation, or how the “checklist” is technically implemented and enforced within an LLM's context window. This lack of technical detail was argued to constitute a failure of both the enablement and written description requirements.

Core Argument for this Ground: Petitioner contended that key claim terms lack objective boundaries, rendering the scope of the claims uncertain. Terms such as “research into a portion of the human-like conversation,” “control signals contribute in part to construction,” and “kickout category” were asserted to be vague and subjective. The petition argued that without clear definitions or structural guidance in the specification, a POSA would have to guess at their meaning, failing the “reasonable certainty” standard established in Nautilus. The functional nature of terms like “control logic” without corresponding structure was also identified as a basis for indefiniteness.

Core Argument for this Ground: Petitioner asserted all claims are invalid under 35 U.S.C. §101.
- Alice Step 1: The claims were argued to be directed to the abstract idea of automating a longstanding human mental process: following a script (checklist), listening to answers, deciding what to ask next, and consulting a reference (second opinion) if unsure. This falls into recognized abstract idea categories of organizing human activity and mental processes.
- Alice Step 2: The claims allegedly lack an inventive concept to transform the abstract idea into a patent-eligible application. The recitation of two LLMs, control logic, and a checklist was framed as the use of generic, conventional computer components to perform their well-known functions. The patent does not disclose any specific improvement to computer technology, such as a new model architecture, a more efficient training or inference process, or a novel data structure.
Additional Grounds: Petitioner asserted §103, §112, and §101 challenges against all dependent claims (2-18 and 20). These arguments contended the additional limitations—such as using LLMs for trigger detection (Claim 2), performing research on prescriptions (Claim 3), or terminating a conversation (Claim 9)—were either obvious over the prior art or recited further abstract steps and functional limitations without enabling disclosure or definite scope.

Petitioner requests that the Board institute a Post-Grant Review and cancel claims 1-20 of Patent 12,142,371 B1 in their entirety as unpatentable under 35 U.S.C. §§ 101, 103, 112(a), and 112(b).