At KyroDB, I worked on context correctness for agents and RAG, including freshness checks that keep stale context out of prompts. I also contributed to KyroBench, a context-retrieval benchmark, and trace ingestion for Smriti, an agent memory runtime.
In my projects, I built systems for GraphRAG, coding-agent orchestration, and recursive language models. My work includes retrieval tracing with 71% agreement and 73% recall, and Wasm sandboxes that used 63% less memory per agent.

