At NVIDIA, I designed a real-time diagnostics framework for GPU and display subsystems, capturing hardware-register state through interrupt-driven RPCs. I also engineered frame-boundary register sampling that reduced GPU and panel-state debugging time from hours to minutes for internal driver engineers.
I built marshalling logic and user-mode debugging APIs and SDKs to expose kernel-mode display-driver functionality and per-frame telemetry. This work supported diagnostics and troubleshooting under strict memory and timing constraints.
On FactLens, I developed a multilingual fake-news analysis application with translation, evidence retrieval, and claim-evidence comparison in a Django REST backend. I also built authentication and AWS S3 media-upload functionality, and on DrugStance Sentinel I integrated multimodal LLM analysis with a real-time dashboard and geospatial map.

