At Nagarro, I design evidence-grounded RAG workflows that answer natural-language questions about internal policies, operational documents, financial statements, regulatory filings, and confidential deal materials with source-linked answers.
I built hybrid retrieval pipelines combining semantic search, BM25, metadata filtering, and cross-encoder ranking. I also developed LangGraph workflows for query routing and answer refinement, with RAGAS evaluation and deterministic financial validation.
At Futurefold Software Pvt. Ltd., I developed an XGBoost model using 7M+ transaction records to predict web sales returns and support proactive inventory decisions. I implemented time-series regression with a reported MAPE of 8% and deployed batch and REST API prediction pipelines.
My work at Futurefold contributed to outcomes including 15% lower excess stock, $500K in annual cost savings, 5% higher revenue retention, and 12% lower storage costs. I also built CI/CD workflows for model training, validation, deployment, and retraining.
