At Defence Research and Development Organisation (DRDO) - SAG, I benchmarked YOLOv8, YOLOv11, and RT-DETR on a 7,255-image drone-vs-bird dataset. YOLOv11 reached 98.7% mAP50, and I built real-time detection pipelines for live surveillance.
I also designed a multi-camera re-identification component to maintain object identity across camera views. I worked with DRDO scientists across evaluation cycles to validate computer vision models against real-world test scenarios.
On ContextLens, I built a code intelligence agent that uses AST-based indexing and hybrid retrieval to generate source-grounded answers with file and line citations. I also developed a retrieval evaluation framework and integrated MCP tools with a Streamlit interface.
I'm working on a predictive maintenance system that estimates Remaining Useful Life from NASA C-MAPSS sensor data. I compare regression, gradient boosting, and deep learning models, and serve the model through a containerized FastAPI endpoint.
