Muhammad M.

I build production RAG systems and AI agents for legal-tech and regulated documents. Recent: a 50,000-document legal RAG platform with Elasticsearch + Claude, 98% multi-jurisdiction anonymisation, sub-2s retrieval and three-tier access control. I'm an AI/ML Engineer and Full-Stack Developer with 4+ years shipping systems past the demo stage into real production use: architecture, backend, testing, deployment, and ongoing reliability (evals, guardrails, monitoring). WHAT I BUILD → RAG & Agentic Retrieval: enterprise document search, legal/compliance platforms, multi-jurisdiction PII anonymization (98% detection accuracy), retrieval pipelines built around evals, not just vector search → Document AI & OCR Pipelines: scanned PDFs and mixed-format records turned into clean, structured, searchable data, with answers that cite their source → AI Agents & Multi-Agent Systems: LangGraph, CrewAI, Claude Agent SDK and Model Context Protocol (MCP) systems with human-in-the-loop approval steps, persistent memory, guardrails, and conversational AWS Bedrock agents → AI-Powered SaaS Backends: Claude/GPT-4o chat with prompt caching (up to 90% token-cost savings), Stripe billing, Twilio/WhatsApp and voice-agent pipelines → Computer Vision & Predictive ML: medical imaging (CNN classification), demand forecasting, statistical prediction models RESULTS I'VE DELIVERED - 80% reduction in document-generation time on a legal AI platform with 50,000+ documents - 98% PII-detection accuracy across multi-jurisdiction legal case data - 95%+ backtested accuracy on real-time prediction models - ~90% input-token cost reduction via prompt caching on a production SaaS chatbot HOW I WORK Discovery call → architecture proposal → scoped build with regular check-ins → testing/evals → deployment & documentation. Clean, maintainable code and a real handoff, not a black box. Tools & stack: Claude API, Claude Agent SDK, OpenAI GPT-4o, LangChain, LangGraph, CrewAI, Model Context Protocol (MCP), Pinecone, Qdrant, Elasticsearch, AWS Bedrock, OCR, Python, FastAPI, Node.js, Stripe, Twilio Top Rated on Upwork with 100% Job Success and a 5.0★ average. If you're building a RAG system, an AI agent, or a document-heavy AI workflow, send me your brief. Happy to do a free 15-minute call to scope it out.