My soccer tracking project combines fine-tuned YOLOv8 detection with re-identification and camera-motion compensation. It reached 84.4% mAP@50 and runs end to end at 87ms per frame.
For OmniMath-Local, I built an async FastAPI backend serving a local model through Ollama, with ChromaDB retrieval and SymPy verification. On an RTX 3050 with 4GB VRAM, average request latency was 2.06 seconds.
I also built AURA HUD, an offline desktop assistant with a Rust/Tauri shell and FastAPI sidecar. It streams audio to Faster-Whisper for offline voice input and uses Ollama for local responses.
As a Software Engineering Intern at Talent Corner HR Services Private Limited, I built Flask REST APIs connecting a Python AI document generator to a React/Node.js frontend and implemented OTP-based login verification.
