ContractScan AI: Legal Clause Auditor

The Problem

Startups and engineering agencies sign dozens of vendor contracts, NDAs, and Master Services Agreements annually. Retaining external legal counsel for routine 15-page standard contracts costs $800–$2,500 per review and introduces 3–5 day deal delays. Crucial risks like unlimited consequential damages, asymmetric IP assignment, or 90-day auto-renewals frequently slip through manual scans.

I engineered ContractScan AI to act as an automated first-pass reviewer: upload a PDF/DOCX agreement, extract contract clauses with tokenized NLP boundaries, and classify risks against benchmarked commercial standards.

Architecture & Semantic Pipeline

The system utilizes a secure, zero-data-retention architecture:

1. Document Parsing & Optical Boundary Detection: Converts PDF/Word contracts into structured text chunks while preserving section numbering and hierarchy.
2. spaCy Tokenization & Clause Classification: Rule-based heuristics and embeddings categorize clauses (Limitation of Liability, Indemnity, Termination, Governing Law, IP Rights).
3. Azure OpenAI GPT-4 Evaluation: Passes categorized clauses through strict JSON schemas to evaluate deviation from balanced commercial standards and compute risk severity (Low, Medium, Critical).
4. Redline Synthesis: Generates replacement contractual language tailored for buyer or vendor protection.

The Stack

Key Capabilities

• Automated Risk Heatmap: Color-coded clause ratings (Red = Critical Uncapped Risk, Amber = Caution, Green = Standard).
• Instant 1-Click Redline Generator: Outputs standard ABA-compliant substitution clauses.
• Governing Law & Jurisdiction Checker: Detects high-friction international jurisdictions and suggests neutral domestic venues (e.g. Ontario/New Brunswick, Canada or Delaware, USA).