Securing Agentic AI

Rolled out code scanning broadly to scan over 95% of the code written at the company from 10%. Wrote CodeQL rules to improve fidelity of findings which resulted in an open-source contribution back to the upstream repository. Implemented a partitioning scheme to scan large monorepos.

Application Security Engineering - developing and maintaining the code scanning pipeline to secure code from first party vulnerabilities. Occasionally also perform security audits based on the code scanning findings. Writing custom rulesets to improve scan accuracy and effectiveness. Developing and maintaining the scanning pipeline for PCI compliance.

Worked on building a private trigger-action platform similar to IFTTT in functionality but which works entirely on encrypted user data. Built a system -- Walnut using MPC and trusted execution environments at a dollar cost of 3.74x over a non-secure baseline. Built another prototype that uses fully homomorphic encryption without any trusted parties instead of MPC.