AI Document Classification Engine for a Legal-Tech SaaS Product

A Hyderabad-based legal-tech SaaS company needed to add an AI-powered document classification and extraction feature to their existing platform. The feature needed to automatically categorise incoming legal documents (contracts, NDAs, court filings, invoices) and extract key clauses for downstream workflow automation. Their internal team had no ML expertise, previous attempts with open-source models had produced accuracy rates below 70%, and the product roadmap deadline was firm.

Nexeratech's AI team proposed a Retrieval-Augmented Generation (RAG) architecture using a fine-tuned language model combined with a structured extraction layer. In the first two weeks we delivered a working prototype connected to a sample document corpus of 500 legal files. We then ran a three-week fine-tuning phase using the client's actual historical document library of 12,000 files. The final production system was built with a FastAPI backend, integrated into the client's existing SaaS platform via REST API, and deployed to AWS Lambda for serverless scaling. Full documentation and model retraining instructions were handed over to the client's team.

The production model achieved 94% classification accuracy and 89% key-clause extraction accuracy across all tested document types — significantly exceeding the client's 85% target. The feature went live six weeks from project start. The client's product team estimates it saves legal-team users an average of 90 minutes per day in manual document review.