Muhammad M.

I build AI/ML solutions based on the problem — not just the technology. Sometimes the right solution is a traditional ML model. Sometimes it's a RAG pipeline or an AI agent. I look at the data and cost before deciding what approach makes sense. Recently, I built a repair-order classification and summarization system for an auto repair business. It processes incoming repair-order text using trained machine learning models instead of relying on per-call LLM APIs, helping keep operating costs predictable as usage grows. What I can help with • Text classification & information extraction — support tickets, repair orders, documents, and other unstructured text using NLP and traditional ML. • Predictive modeling — classification, regression, feature engineering, and model evaluation using scikit-learn, XGBoost, and other ML approaches. • RAG applications — question-answering and retrieval over private documents using LangChain, FAISS/ChromaDB, and models from OpenAI, Claude, or Ollama. • AI agents & workflow automation — LangGraph-based agents, API/MCP integrations, and n8n workflows for automating multi-step tasks. • Computer vision — object detection and segmentation with YOLO and PyTorch, including dataset preparation, training, and evaluation. • Data analysis & ML on real-world data — cleaning messy datasets, feature engineering, exploratory analysis, and evaluating models properly. I've also worked on a confidential computer-vision project involving dental imaging, as well as applied ML work involving real-world scientific and literature-review data. I can walk through the approach and technical details where the work can be discussed. Tech I work with: Python • scikit-learn • XGBoost • pandas • NumPy • PyTorch • TensorFlow • OpenCV • YOLO • LangChain • LangGraph • FAISS/ChromaDB • Hugging Face • OpenAI • Claude • Ollama • FastAPI • Docker • n8n • SQL If you're unsure whether your project needs a traditional ML model, RAG system, or AI agent, send me the problem and what you're trying to achieve. I'll help you determine which approach makes sense before jumping into implementation.