Sushant Dagar

With LedgerMatch, I built AI document parsing that pairs OCR and an LLM with deterministic validation. Python checks arithmetic relationships and source text, while mismatches in purchase-order, goods-receipt, and invoice matching go to logged human approval.

I built a document Q&A service with Flask, ChromaDB, and Gemini embeddings, then containerised and deployed it on AWS EC2. I also built an evaluation harness for a LangGraph agent that scores routing, retrieval, and answer faithfulness as a regression gate for model and prompt changes.

For a turbofan remaining-useful-life project, I trained an LSTM and Random Forest on NASA C-MAPSS data, reducing test RMSE from 44.7 to 13.6 cycles. I'm a B.Tech IT student at IIIT Una and an AWS Certified AI Practitioner, available for a six-month full-time internship from January 2027.