BCAQI Labs
日本語 Book a 15-minute introduction
Work

See a working system before you talk to us.

Below is a live demonstration of a Japanese document intelligence pipeline, followed by case studies from our research and engineering work.

Japanese Document Intelligence Pipeline

Pick a preloaded sample below. The pipeline extracts fields, normalizes Japanese dates (和暦 → ISO), validates against business rules such as the 10% consumption tax, and routes anything uncertain to a human-review queue. This is exactly the class of work that a general chat assistant cannot do reliably.

Document Intelligence Pipeline Demonstration environment

Sample documents

Samples contain no real personal data. Nothing you upload leaves your browser in this demo.

Select a sample document to run the pipeline.

This is a demonstration with representative output. In a paid pilot we run your own documents and layouts through the production pipeline, measured against an agreed accuracy target.

Case studies

Research-grade engineering, applied.

Featured at ICML 2025

AutoML-Agent

Problem
Building a machine-learning pipeline — data prep, model selection, tuning, evaluation — takes expert time and iteration most teams cannot spare.
Approach
An agentic system that turns a natural-language task into a plan, then executes it: multiple specialized agents build, run, and evaluate candidate pipelines under a controller.
Outcome
An end-to-end ML pipeline produced from a task description, with the work presented at ICML 2025.

Architecture (simplified)

NL task
→
Planner agent
→
Data agent
Model agent
Tuning agent
→
Evaluator
→
Pipeline + report
View project ↗
Research collaboration

AI Paper Reviewer

Problem
Reviewing research papers consistently and thoroughly is slow, and quality varies between reviewers.
Approach
A multi-agent system that reads a paper and produces structured, criterion-based feedback — methodology, novelty, clarity, and evidence — rather than a single opaque score.
Outcome
Structured review feedback aligned with how human reviewers assess work.

Architecture (simplified)

Paper
→
Section parser
→
Methodology
Novelty
Clarity
→
Structured review
View project ↗

Additional production work is under NDA and described on request. We never name a client without written permission.

Want this run on your documents?

A 4-week feasibility PoC turns this demo into a measured result on your own data.

Book a 15-minute introduction See engagement models