Software engineer — LLM/RL research, autonomous robotics, and production full-stack systems.
Merged code into the production codebases of Meta, Google, OpenAI, and Hugging Face.
Meta × Scaler PyTorch Hackathon Grand Finalist · Top 30 nationally, ISRO Robotics Challenge (IRoC-U 2026)
| Org | Repo | PR | What it fixed |
|---|---|---|---|
| Meta | facebookresearch/hydra | #3262 | Fire on_job_end/on_run_end callbacks on KeyboardInterrupt |
| google/osv-scanner | #2762 | Skip malformed commit hashes instead of aborting the scan (+711 LOC, Go) | |
| OpenAI | openai/openai-agents-python | #2931 | Surface run-loop exceptions after stream_events() completes |
| Hugging Face | huggingface/transformers | #46006 | Fix /v1/models response type in transformers-serve |
| SchemaStore | SchemaStore/schemastore | ×3 | TypeScript/JSConfig lib entries + Claude Code settings schema |
- promptinject-env-agents — adversarial multi-agent RL (GRPO) for LLM prompt-injection defense; defender accuracy 47.1% → 92.8% over 30 A100 episodes. Live on Hugging Face Spaces.
- MindRead-RL-Environment — Theory-of-Mind RL environment, also live on HF Spaces.
- Autonomous GNSS-denied micro-UAV — ISRO Robotics Challenge (IRoC-U 2026), Top 30 nationally. Custom visual-inertial SLAM, ArUco precision landing, RealSense + YOLOv8 tracking on a Raspberry Pi companion computer.
- Keepi — local-first AI memory app on Flutter/Isar with a realtime Gemini Live voice assistant, live on Google Play.