At Flipkart (Shopsy), I built and deployed a Flask application where an LLM conducts multi-step investigations using conditional tool calls. I designed its system prompt, tool descriptions, and dialogue flow, cutting analysis time by ~60%.
I also built a 10,000+ query evaluation benchmark that drove a 5% relevance gain, and wrote structured-output rubric prompts that contributed to an 18% increase in autosuggest adoption. Earlier, at BNP Paribas, I supported banking application backends and production troubleshooting.

