Insurance SaaS Platform — Engineering Case Study

LLM/OCR Project · by Yatharth Lakhera

Built AI-powered insurance processing system leveraging LLM/RAG for document classification, compliance detection, and data extraction.

The challenge

Insurance teams were spending ~5 hours per claim manually classifying documents, cross-referencing policy data, and flagging compliance issues. Throughput was capped by human review, and a single misclassification could cascade into denied claims and audit risk.

The build

Built an end-to-end LLM/RAG processing pipeline using Document AI, Gemini, and ChatGPT for classification and extraction. Layered a custom search index for cross-referencing and ran compliance detection automatically at upload time — surfacing issues before a human ever opened the document.

Impact

Processing time dropped from 5 hours to 1 hour — 80% faster
95% classification accuracy across heterogeneous insurance documents
Compliance issues flagged at upload, before they hit the review queue

Technologies: ChatGPT, DocumentAI, Gemini, OCR, Python, FastAPI

Yatharth Lakhera — full portfolio