At Turing, I create prompts, adversarial tasks, and golden reference solutions for SFT, RLHF, and LLM benchmarking. I also build Python tooling and visualization notebooks to support training, evaluation, and behavior analysis.
I design Agentic AI workflows using ReAct and Plan-and-Execute, with domain-specific API replicas for multi-step reasoning and tool calling. I also contribute to tool design, SOP development, and quality-assurance workflows.
For an education chatbot serving 5K+ students, I implemented RAG enhancements through chunking and retrieval tuning to improve textbook-grounded extraction. I also led a 10-member pod, overseeing task allocation, quality reviews, workflow alignment, and performance tracking.
Previously, at Code Clause, I analyzed 0.5M+ EV battery records and developed an LSTM model that improved range prediction accuracy by 18%. My projects include a Python-based MCP server, a multi-agent solar feasibility system, and a web-based LLM research assistant.

