LegalTech · AI Contract Intelligence
Finding the Clause in Ten Seconds Instead of Ten Days, in Japanese and English
Project Snapshot
- Client : ConPass AI, a next-generation contract analysis platform
- Industry : LegalTech and enterprise services
- Scope : Unified contract workspace, semantic search, AI risk detection, bilingual understanding
- Platforms : Web
- Technologies : Python (FastAPI), React, MySQL, Redis, Google Cloud Platform, Gemini Cloud Assist, LLMs with RAG, Cursor
- Languages : Japanese and English contracts, including 甲/乙 party structures
The Impact
10×
Faster contract discovery and review compared with manual search
85%+
Precision on AI-driven risk identification
0
Missed exposure, with renewals, deadlines and obligations surfaced early
JP / EN
Bilingual understanding of contract structure, parties and metadata
Overview
ConPass AI serves organisations holding enormous volumes of legal and operational agreements — the kind of repository where the answer to “are we exposed here?” exists somewhere, and nobody can find it in time. Three groups were carrying that problem:
- legal and compliance teams, reviewing by hand and deciding late,
- operations, working across contracts scattered over several storage platforms, and
- bilingual organisations, where Japanese and English agreements follow different structural conventions.
The bilingual point is what separates this from a generic document-search build. A Japanese contract identifies parties as 甲 and 乙 rather than by name, and a system that cannot interpret that convention cannot answer even the simplest question about who owes what to whom.
Brain Station 23 partnered with ConPass AI to engineer an AI-first platform that centralises contract intelligence: one workspace, natural-language search over all of it, and automated detection of legal, financial and operational risk.
The result turns a repository people were afraid of into something they can interrogate, in either language, before a deadline passes rather than after.
Business Challenges
Contract risk is rarely a mystery. It is a known clause in a known document that nobody re-read at the right moment. Every pressure here is a version of that.
Five pressures shaped the platform.
- Data was fragmented. Contract data sat across multiple storage platforms with no unified visibility, so the first step of any review was finding the document.
- Review was a manual bottleneck. Heavy reliance on human review produced long turnaround times and delayed decisions.
- Compliance exposure was real. Missed obligations, expirations, renewals and regulatory gaps were a standing risk in a repository nobody could fully survey.
- Bilingual contracts broke ordinary tooling. Processing Japanese and English agreements, and interpreting structures like 甲/乙, defeated systems built for one language.
- Risk could not be detected at scale. There was no way to quickly identify or predict financial and operational risk across vast document volumes.
Together they explain why “just search it” was never an option: search without structure, language awareness and risk logic returns documents, not answers.
Discovering renewals after they pass?
Bring it to a 30-minute call — we will come back with an architecture and a scoped estimate.
What We Built
The build follows the order in which a contract question actually gets answered: find everything, understand it in its own language, then let the system read for risk before a person has to.
A unified contract workspace
One place to store, search and view every agreement, which removes the discovery problem before any intelligence is applied.
Natural-language search
Users ask plain questions and get clauses back, built on retrieval-augmented generation so answers are grounded in the actual contract text rather than a model’s guess.
AI risk insight
Legal, financial and operational risks are identified and explained clearly, with the 85%+ precision that makes the output trustworthy enough to act on.
Bilingual understanding
Japanese and English structures and metadata are interpreted smoothly, including party conventions like 甲/乙, so the same question works in either language.
Smart filters
Parties, dates, categories and contract attributes are auto-detected, giving teams structured navigation over what used to be an unstructured pile.
Zero missed obligations
Renewals, deadlines and obligations are identified early and surfaced as alerts, turning compliance from a periodic audit into a continuous signal.
Secure, permission-based access
Contract visibility is governed by permissions across teams, so centralising the data does not mean everyone can see everything.
The Results
The gains land where the pain was: how fast a question gets answered, how much is caught, and how much of the load still falls on manual review.
Speed and precision
Contract discovery and review run 10× faster than before, and AI-driven risk identification reaches 85%+ precision. Automated risk alerts mean zero missed exposure on the obligations, renewals and deadlines that previously depended on someone remembering to look.
Governance and leakage
Compliance and governance are stronger because the whole repository is visible and continuously read. Dependence on manual contract review drops, and with obligations caught early, revenue leakage and operational risk both fall — the financial cost of the clause nobody re-read.
The transferable lesson is that the language work was not a localisation task. Handling 甲/乙 correctly is the difference between a system that can answer “what are our obligations to this party” and one that cannot, and it had to be designed in rather than bolted on.
The other is grounding. Retrieval-augmented generation is what lets a legal team trust a generated answer, because every answer points back to the clause it came from.
Holding contracts you cannot search, in more than one language?
You might be reviewing by hand, discovering renewals after they pass, or running tooling that cannot read half your repository. In every case, the first step is the same.
Book a 30-minute call with Brain Station 23. Bring your document volume and the languages it comes in, and we will come back with a proposed architecture, a delivery plan and a scoped estimate.
Discovering renewals after they pass?
Bring it to a 30-minute call — we will come back with an architecture and a scoped estimate.
Related Case Studies
AI & ML
GiologiQ: Reducing Well-Drilling Risk with Real-Time Digital Twin and AI Visualization