Muhammad Umar A.

I build computer vision systems that run in production, not just in notebooks. From real-time detection and tracking on CCTV and drone footage to optimized edge inference on NVIDIA Jetson, I take projects from raw data and annotation through training, optimization and deployment. When a project also needs a language layer (RAG, agents, document AI), I build that too. 🔍 COMPUTER VISION (my core focus) • Object detection, tracking & segmentation: YOLOv8, Faster R-CNN, DeepSORT, ByteTrack • Pose estimation, OCR / Document AI, medical imaging, industrial inspection, anomaly detection • Edge deployment on NVIDIA Jetson with TensorRT, ONNX, INT8/FP16 quantization • Custom dataset creation, annotation workflows and model evaluation (mAP, precision/recall) 🧠 LLMs, RAG & AI AGENTS • RAG pipelines with LangChain, LlamaIndex and vector databases (Pinecone, Qdrant, ChromaDB) • Fine-tuning and prompt engineering (OpenAI, Claude, Llama, open-source models) • AI agents and multi-agent systems (LangGraph, CrewAI, function calling) • Multimodal apps that combine vision and language models ⚙️ DEPLOYMENT & AUTOMATION • Real-time inference APIs with FastAPI/Flask, Docker, Kubernetes, CI/CD, AWS & GCP • MLflow experiment tracking, scalable cloud + edge architectures • Workflow automation with n8n, Make & Zapier; chatbots and voice AI for web, WhatsApp and Telegram INDUSTRIES: retail analytics, surveillance, drones, healthcare, industrial automation. HOW I WORK • Clear scoping up front: success metrics, dataset needs and deployment target agreed before I write code • Regular updates and honest communication, including early warning when something won't work • Clean, documented code and a handover your team can maintain • Systems built for real-world conditions: latency, hardware limits, messy data 📩 Send me a short description of your project (what you want to detect or automate, your data, and where it needs to run) and I'll reply with an approach and timeline.