Skip to content
Projects

AI Contract Review System

End-to-end AI system for contract analysis covering extraction, checklist review, risk assessment, and revision suggestions. Patent registered.

2023.09 — Present BHSN AI Engineer
Legal AI NLP Contract Analysis FastAPI Patent

Overview

Built and evolved a production AI system for automated contract review. Covers the full pipeline from document extraction to checklist-based review, risk assessment, and revision suggestions, processing multilingual contracts (Korean, English, Japanese).

Core Features

Contract Analysis Pipeline

  • Extraction: Clause-level extraction, party identification (97.5% accuracy), contract classification
  • Checklist Review: Custom playbook-based AI review (Green/Yellow/Red risk grades) — evolved from per-clause V1 to full-contract V2 API
  • Summarization: Structured contract summaries with key term identification
  • Revision Suggestions: Context-aware clause revision recommendations (91-93% fix ratio on production contracts)

Playbook Self-Serve Platform

  • Designed a platform enabling legal professionals to independently author, test, and deploy review playbooks via Excel-based workflows
  • Client-specific playbook overlay structure — maintaining shared playbooks while applying workspace-level customizations
  • Automated playbook release pipeline: contract type SQL, playbook build, warning triage, handoff

Multi-Editor Integration

  • Extended AI review for direct use within document editors (Hancom HWP add-on, Word plugin)
  • Led API specification design for external partner integrations

Scale

  • 2.5 years of continuous development
  • Multilingual support: Korean, English, Japanese
  • 38 contract types, 4 client PoC deployments

Patent

Registration No. 10-2867168 (Republic of Korea) — Registered patent for AI contract review methodology. Covers contract analysis, summarization, checklist review, and model training.

Academic Context

The retrieval pipeline for contract analysis follows the Advanced RAG paradigm, and structured output generation applies JSON schema validation in the context of constrained decoding research such as XGrammar (Dong et al., 2024).

While the foundational CUAD (NeurIPS 2021) dataset covers 13K annotations on single-language contracts, this system extends to 38 contract types across Korean, English, and Japanese with client-specific custom playbook overlays. This system achieves production-grade performance through domain-specific fine-tuning and a playbook-based approach, and includes a citation verification system and related clause linking.

Impact

Operating as a core product feature enabling legal teams to review contracts more efficiently with AI-assisted analysis.