At Runwall, I built a zero-trust proxy between AI agents and their tools over MCP, enforcing OPA/Rego policy on every tool call. Four adversarial security audits raised the security score from 2.1/10 to 9.1/10, and the launch reached 3k+ monthly unique visitors and 30k+ requests.
At DRDO, I analyzed flight data recorder data and trained models for event classification and sequential risk prediction, improving accuracy from 82% to 93%. At Tellis Technologies Pvt Ltd, I implemented RAG systems and finetuned language models to reduce hallucinations; I’ve also contributed security and reliability fixes to Superagent and LiteLLM.
