Built the managed agent harness - the core infrastructure that runs AI agents in production at AWS scale. Implemented the full agentic loop end-to-end: model reasoning, tool selection, action execution, and response streaming. Engineered stateful, isolated microVM sessions with persistent filesystem and shell access, automatic memory sanitization between sessions, and zero cross-session data leakage. Built in multi-model support (Bedrock, OpenAI, Gemini) with mid-session provider switching, MCP tool connectivity, and automatic CloudWatch observability - all configurable with no orchestration code on the caller's side.

Worked on S3 Space Reclamation team, owning the core S3 delete engine that processes 300PB per day. Conducted zero-downtime rollout of updated delete engine (replacing 20-year legacy system), preventing capacity crunch worldwide and saving Amazon billions annually.

At Codexis, a global leader in enzyme engineering, I conceived, designed, and built several Machine learning tools for directed evolution of biotherapeutic protein projects:

• An LSTM based Post-translational Modification prediction tool for gene therapy (industry-leading 0.968 MCC)
• An Antimicrobial Peptide identification tool (> 0.95 F1 score)

Relevant Coursework: Programming in C, Electrical Engineering, EE Lab, Networks, Machine Learning, Theory of Computation, Computer Architecture, Linear Algebra and Applications, Probability and Statistics, Differential Equations, Multivariate Calculus