AI and backend engineer. Agents, pipelines, and the systems around them.

Agent bypassed its tools and called the GitHub API directly using a broadly scoped token. Proposed and implemented two-token architecture - agent gets narrow permissions, resolver keeps broad ones.

Agentic document search. ReAct loop that reasons about what to read instead of relying on vector similarity. Small embedding model (90MB), agent reasoning compensates. 50% latency reduction over naive RAG.

Fine-tuned DeBERTa on ~51k examples for prompt injection detection. Deployed to Hugging Face in 16-bit and 8-bit.