Legal contract intelligence
Reading M&A contracts five times faster
- Year
- 2020–24
- Client
- SourceFuse
- Domain
- Legal · GDPR
- My role
- AI/ML Technical Lead
- Stack
5×faster contract review
92%F1 on clause extraction
1M+API requests a day
The problem
Legal teams reviewing M&A deals had to read every contract for indemnities, liability caps and termination clauses — slow, expensive and easy to miss something.
What I built
- Fine-tuned DeBERTa and RoBERTa models for clause extraction and contract-type classification (89% accuracy).
- Hybrid retrieval — FAISS dense, BM25 sparse and a cross-encoder reranker — with 40% lower latency and 34% better MRR.
- An OCR pipeline for 50,000+ scanned contracts a month, deployed as auto-scaling microservices.
- SHAP and LIME explanations so lawyers could see why the model flagged a clause.
The hard part
Lawyers won’t trust a black box. Pairing every extraction with an explanation turned the model from something to double-check into something to work alongside.
Outcome
Contract review got 5× faster while staying GDPR-compliant and explainable.
Client data and code are confidential; figures are from the delivered system.
Next caseCivic data agents →