Harshit G M User

At Novatechrobo, I own the AI/ML stack for an autonomous service robot, taking computer vision, edge inference, and real-time control integration from prototype to a live preschool pilot deployment.

I build real-time face tracking, object detection, and gesture recognition pipelines using Python, OpenCV, and YOLOv8/v10. I also engineered ESP32-to-edge-compute communication that cut end-to-end command latency by 30% and coordinated encoder and IMU sensor fusion for live-hardware validation.

In my Savage Sentinel project, I designed a LangGraph-based agentic RAG copilot with hybrid retrieval, cross-encoder reranking, self-healing tool execution, Redis semantic caching, and automated RAGAS regression evaluation. The semantic cache reduced repeat-query latency from roughly 1,800ms to roughly 12ms.

I enjoy building measurable AI systems that work beyond offline demos, from LLM-powered retrieval products to perception and control stacks on constrained hardware.