I built QueryMind, a natural-language BI platform that turns plain-English questions into validated, read-only SQL, explanations, charts, and saved dashboards. Its 12-tool agent validates and safely executes queries, with about 30 test files and a hardened GitHub Actions CI pipeline across 75+ merged pull requests.
At Arch Technologies, I built RAG pipelines and agentic workflows in Python and Docker for document-based AI applications using real client datasets. I also developed evaluation workflows to benchmark retrieval quality, LLM reliability, and hallucination risk, then used the measurements to tune prompts and retrieval logic.
I’ve also built a document-Q&A RAG backend with hybrid retrieval and citation-backed answers, and a Shopify store-operations agent with human approval for AI-initiated actions. Across these projects, I focus on making LLM workflows testable, traceable, and safe to run.

