At Mercor, I engineered competitive programming problems, solutions, and edge-case test suites, then validated them in containerized Docker environments and GitHub CI workflows.
I also designed AI prompts, evaluation rubrics, and golden reference solutions in Airtable to benchmark LLM reasoning. I analyzed software engineering traces and reviewed autonomous AI coding agents using Cursor, Claude Code, and Specystory.
I built SafeGuard, a context-aware AI surveillance system, and fine-tuned YOLO11s on a custom annotated weapon dataset. It achieved 92.0% mAP@0.5, while asynchronous CUDA streams and parallel GPU execution reduced end-to-end latency by 85%.
For Acharya AI Fitness Trainer, I built a full-stack platform that generated personalized workout and diet plans and delivered conversational voice coaching. My SafeGuard research was published in IEEE Xplore at CISCT 2026.
