At AccionLabs, I architected and built a multi-agent framework that modernized a 400K+ LOC insurance COBOL estate into .NET and React. It reduced effort per screen from 40–50 days to 2–3 days, with a five-stage agent workflow and a Neo4j knowledge graph to ground code generation in program dependencies.
I also re-engineered a live hybrid search and recommendation engine for nearly 40 million products on AWS. By using field-level embeddings and hybrid reranking, I brought product-ID search accuracy to nearly 100% and surfaced the correct product in the top two results.
At AccionLabs, I built a Tableau-to-Power BI migration pipeline that reduced work per report from 1–2 days to 5–10 minutes, with 95% visual accuracy across 10 validated reports. I also developed an agentic RAG voice assistant that combines retrieval and tool-calling agents to produce grounded responses.
Earlier, as an AI/ML Developer intern at Upperthrust Technologies, I built a computer-vision pipeline for a live mobile urinalysis app and delivered its model for on-device inference. I also developed an OCR-to-JSON pipeline that routes non-conforming fields for human review.
