I built an LLM evaluation harness that runs test sets through RAG retrieval and generation, scoring hit rate, MRR, precision@k, and answer quality. It uses Groq as an independent judge and tracks latency and failure modes to flag regressions.
For my AI Code Reviewer project, I designed a structured prompt for Groq-hosted LLaMA 3.3 70B and built FastAPI endpoints with Pydantic validation and error handling. I deployed it on Azure App Service.
I also built a personal research agent using LangGraph’s ReAct loop, Gemini, and Tavily tool calling. I debugged PDF export issues and added Quick Summary and Detailed modes.
As a Data Science Intern at NexGen Analytix, I built ETL workflows and an employee attrition prediction model. I also wrote SQL queries to segment at-risk groups and built a Power BI dashboard.
