I've built production AI systems at Corazor Technology and Ethaa Private Limited, from real-time driver safety monitoring to reliable LLM agent workflows.
At Corazor Technology, I architected a driver monitoring pipeline using MediaPipe, OpenCV, solvePnP, FastAPI, and WebSocket, achieving sub-50ms inference per frame. I combined EAR, MAR, PERCLOS, blink-rate logic, LSTM smoothing, and hysteresis to reduce false-positive fatigue and distraction alerts by approximately 35%.
At Ethaa, I optimized prompts for OpenAI GPT, Anthropic Claude, and Mistral, while building validation frameworks for tool calls that improved agent reliability by over 40%.
I've also built a multi-agent QA automation framework that generates and executes API tests, and implemented GPT-2 from scratch in PyTorch. I bring hands-on experience across LLMs, agentic AI, computer vision, MLOps, and cloud AI platforms, backed by Kaggle competition work and 600+ algorithmic problems solved.
