EQUES Inc., an AI startup originating from the University of Tokyo’s Matsuo Laboratory, has equipped its AI document checking system for pharmaceutical quality assurance with new features that prioritize detection results and highlight source evidence. Behind this is a strong industry demand to further reduce the burden of post-detection human review in the rigorous document consistency checks required in pharmaceutical manufacturing.

Key Points of the Event and New Features of QAI Checker

On August 31, 2026, EQUES Inc., an AI startup originating from the Matsuo Laboratory at the University of Tokyo, rolled out a major update to its “QAI Checker” AI document checking system for pharmaceutical quality assurance, releasing three new key features. The features added through this update are the “Evidence Location Marker,” “Discrepancy Score,” and “Discrepancy Inspection Agent Version,” which aim to further reduce the time required to review and make decisions on detection results compared to previous features. In pharmaceutical manufacturing sites, it is strictly mandatory that descriptions are consistent across numerous complex documents, such as marketing authorization applications submitted to the government and manufacturing procedures used on-site. Traditionally, “QAI Checker” has analyzed document layouts to automatically match paragraphs with the same meaning, detecting inconsistencies such as numerical discrepancies and omissions in processes. This latest feature addition was developed to powerfully alleviate the burden in the “post-detection review” process, where personnel interpret where the extracted inconsistencies need to be corrected following such automatic detection.

“Evidence Markers” and “Discrepancy Scores” to Optimize Review Operations

Among the new features, the “Evidence Marker” is a feature that directly highlights the descriptions on the document that served as the basis for the AI’s determination of an inconsistency. This allows users to understand at a glance which descriptions in the approval document and the procedure manual were compared to cause the discrepancy. While traditional review work involved the manual effort of repeatedly comparing the original texts to find the cause, this feature significantly reduces that burden. In addition, because the highlighted screen can be saved as-is, the accountability and transparency of the review records are also improved. Please refer to the interface image below.

. Another new feature, the “discrepancy score,” presents the possibility of inconsistencies extracted by the AI for each section as a percentage or numerical score. This allows reviewers to clearly set priorities on which feedback to check first, enabling them to reliably address critical errors of particularly high importance within limited working hours.

Technical Background and Approach to Pharma-Specific Models

The Evolved Agent-Based Approach: “Discrepancy Check Agent Edition”

The third key feature, the “Discrepancy Check Agent Version,” adopts an AI agent approach and has been fundamentally redesigned to flexibly accommodate a wider variety of document formats, complex descriptions, and different input formats. It incorporates the latest knowledge and algorithms regarding matching accuracy improvement and advanced natural language processing that EQUES has cultivated through its internal cutting-edge R&D processes. Even when comparing documents with irregular paragraph structures—which were difficult for conventional systems to analyze automatically in terms of structure—and unique expression patterns, it is now able to detect discrepancies with high accuracy, significantly expanding the scope of applicable tasks. The overall system architecture is shown in the figure below.

.Since its establishment in February 2022, as a professional implementing cutting-edge machine learning technology into society, EQUES has made its core strength the speed at which it delivers its proprietary solutions to business frontlines within the AI field, where technological updates are exceptionally rapid.

The existence of “JPharmatron-7B”, adopted by the Ministry of Economy, Trade and Industry’s GENIAC

An important technical foundation supporting EQUES’s pharmaceutical AI business is “JPharmatron-7B,” a large language model for the pharmaceutical industry independently developed by the company. Despite its relatively lightweight structure of 7 billion parameters, this model has been trained on large-scale, specialized datasets in the fields of pharmacology and pharmaceutics. In “JPharmaBench,” a proprietary benchmark designed for practical scenarios such as the Japanese National Examination for Pharmacists and discrepancy checks, it has recorded top-tier performance that outperforms existing open-source models of the same scale across all evaluation metrics. Furthermore, this development has been adopted and advanced as part of “GENIAC,” a domestic advanced generative AI development support project sponsored by the Ministry of Economy, Trade and Industry (METI) and NEDO. Because of its lightweight nature, it has the major advantage of allowing pharmaceutical companies—which handle highly confidential information—to easily deploy and operate it within their internal networks or secure local server environments, and it also closely contributes to the intelligent behavior of the various functions integrated into the QAI Checker.

Significance of Practical Application in the Pharmaceutical Industry and Future Prospects

The Practicality of the “QAI Series” That Clears Strict Regulations

In the pharmaceutical industry’s quality assurance operations, even minor notation errors or discrepancies in descriptions can directly lead to serious regulatory risks and product recalls. At the same time, major challenges include a severe shortage of QA personnel—who require specialized knowledge and extensive experience—and the ever-increasing burden of document creation. The “QAI Series” provided by EQUES complies with strict rules such as GMP (Good Manufacturing Practice for pharmaceuticals) through both the AI SaaS “QAI Generator” for automated document generation and the newly introduced “QAI Checker”. These products are developed not merely by introducing standalone AI models, but with consideration to allow humans to easily intervene in the actual operational workflows of pharmaceutical sites. For example, the “QAI Generator” has added automated generation functions for deviation reports and annual reviews, boasting a track record of achieving about a 70% reduction in workload for certain document generation tasks. Regarding the “QAI Checker,” a two-week free trial is also available so that users can verify its accuracy and usability in actual operational workflows in advance.

The Real-World Deployment of AI Expanding from Pharmaceuticals to Critical Infrastructure

While headquartered in Hongo, Bunkyo-ku, Tokyo, EQUES officially opened its Hokkaido branch on August 17, 2026, within “HooK,” a startup support facility in Sapporo, thereby accelerating its efforts to drive industrial DX in regional areas. Looking ahead, the company plans to further enhance the functionality of its pharmaceutical-industry products and solidly establish a comprehensive support system spanning from document creation to inspection. Furthermore, the technological social implementation approach cultivated in this extreme domain of the pharmaceutical sector—where exceptionally high reliability and safety are demanded—is beginning to be horizontally deployed across other critical infrastructure industries as well. Specifically, in May 2026, the company fully launched its “Energy AI Business,” centered on the nuclear power industry, and began offering “Explainable AI” to prevent decision-making processes from becoming black boxes in order to meet audit requirements, alongside autonomously operating “Physical AI.” The world is watching the company’s continuous challenge to seamlessly connect research with frontline operations.

Reference Page

• [EQUES Inc.] https://eques.co.jp/
• [QAI Checker Service Site] https://qai-checker.eques.co.jp/
• [JPharmatron-7B (HuggingFace)]https://huggingface.co/EQUES/JPharmatron-7B

【#EQUES #QAIChecker #PharmaAI #LegalDX #UTokyoMatsuoLab #AIAgents #GenerativeAI】