Shivratna KPS

Built real-time computer vision pipelines for digital rehabilitation, including body keypoint detection and pose estimation for biomechanical motion analysis. Optimized inference services achieving <100 ms predictions for real-time patient motion tracking. Designed scalable AI services with Python APIs and containerized deployment, in collaboration with cross-functional engineering and research teams.

Architected Python-based data ingestion and streaming workflows integrating PyTorch and TensorFlow inference pipelines with a Flutter mobile application. Designed multimodal AI feedback systems combining computer vision outputs with LLaMA 2 reasoning. Optimized video compression and frame alignment, reducing processing latency by 30%. Built scalable FastAPI backend services with asynchronous pipelines handling real-time AI inference workloads.