At Intuit, I designed and built the Evals system from scratch: a schema-based LLM evaluation pipeline now adopted across multiple product lines. It caught a critical widget-description regression before production.
I also fixed an MCP backward-compatibility blocker that enabled ChatGPT and Claude integration into Intuit products. For TurboTax, I own Prompt and Context Engineering, including tool-description metadata for production responses.
At Intuit, I designed tool chaining with server-side invocation caching and LLM instruction guardrails for accurate, ordered agentic workflows. I also researched agentic browsers and an AI Mode on Google track, including a desktop-app-to-MCP-server proof of concept.
Earlier, I built software for ProSeries, Philips X-ray field service, and Carl Zeiss clinical devices, and contributed to Siemens' IEC 61850 System Configurator. I’ve also helped train engineers through Intuit's AAA program and mentor M.Tech AI/ML students at BITS Pilani WILP.

