Khadim H.

AI Engineer | AI Agents, LLM & RAG, Computer Vision, Production ML

I build AI agents and LLM systems that hold up in production: agentic workflows, RAG, MCP servers, and n8n automation. I also do computer vision, from object detection and OCR to real-time video. Top Rated Plus, 100% job success, $250K+ earned. WHAT I DO AI agents and LLM systems. Agentic AI workflows, tool calling, MCP servers, and RAG that cites its sources instead of inventing them. Vector databases, embeddings, prompt engineering, and fine-tuning with LoRA or QLoRA when a smaller model does the job for less money. Usually LangChain or LangGraph with the OpenAI API, Claude API, or Ollama when it has to stay local. A lot of my recent work is agent-assisted dev tooling: Claude Code, OpenAI Codex, Cursor, Copilot. AI automation and chatbots. AI automation with n8n, API integration, ETL and data extraction, plus chatbot development for support and internal tools. Python and FastAPI on the backend, React or Next.js on the front when a project needs a face. Computer vision. Object detection and tracking, image segmentation, anomaly and defect detection, OCR and document extraction, image processing, real-time video. YOLO, OpenCV, PyTorch, TensorFlow. Making these run fast on your hardware is ONNX, TensorRT and Triton work: model optimization, quantization and GPU inference. Machine learning and deployment. Deep learning model training, evaluation and MLOps. Docker, Kubernetes, CI/CD, Linux, GCP, Azure, AWS, and on-prem or edge GPUs. I've spent a lot of time on the unglamorous part, where a model that behaved perfectly on a laptop has to survive a real environment. SOME RESULTS Cut inspection time by roughly 80% with automated visual defect detection for a manufacturer. Took a production inspection model from about 70% accuracy to 99%+. Made inference around 70% faster on the same GPUs using TensorRT, ONNX and Triton. Built RAG assistants over private company data that answer with citations. Speech recognition and transcription pipelines for media workloads. HOW I WORK I move fast, I write code the next person can maintain, and I'll tell you early if I think a target isn't realistic. Not everyone wants to hear that. It still beats finding out in week three. Send me what you're building and what you'd count as finished, and I'll tell you how I'd approach it and roughly what it takes.