Cloud AI Research - LLM Agent safety thru verified code generation.
1. Architected a Secure ADK Plugin (SecureAgent) that connects to VeriGuard Cloud Run service to "compile" natural language security requirements into verified, executable Python policies on startup.
2. Engineered a sandboxed execution environment to mitigate malicious code policy injection and implemented context-aware guardrails to prevent cross-tool data leakage
3. Enabled granular control by injecting security hooks at every stage of the agent lifecycle (pre/post-tool execution and pre/post-model invocation)
4. Implemented a Runtime Tool Wrapper that intercepts specific actions (for e.g. apply_discount), successfully demonstrating multiple flows.
Cloud Security team - Identity and Infrastructure
1. Designed and executed Lifecycle management of GCP resources across all GCP services.
2. Led the project for moving away from the legacy account management system (having Google wide dependency) with clear goals and impact on the OKR to a new distributed ID management service (designed and developed by me) ensuring high reliability.