At GSI Technology, I contributed to a U.S. DoD-funded Air Force Research Laboratory SBIR program developing object detection and segmentation capabilities for the Gemini-II APU. I also built a reproducible benchmarking pipeline comparing computer vision models across the APU, CPUs, and NVIDIA GPUs.
I optimized YOLO variants on real and synthetic image datasets using quantization and mixed-precision training, achieving lower per-image latency than the CPU baseline. I also diagnosed unsupported operations and worked with hardware teams on fixes.
At TIFIN AMP, I built predictive models for advisor purchase likelihood using more than 5 million behavioral and transactional records, improving precision@k by 18% over baseline. I also developed ETL pipelines and a proof-of-concept LLM summary generator for financial advisor insights.

