Muhammad M.

I build multi-agent systems and RAG pipelines that survive production, and I get called in to fix the ones that don't. Agents that fall over under real load. Retrieval that confidently returns the wrong answer. Inference bills nobody can explain. That is most of my work, and it runs on Google ADK and Vertex AI, LangGraph, and AWS Bedrock. Then there is the part almost nobody does. Most AI engineers can't touch an ERP, and most ERP consultants can't build AI. I live in the overlap. I take AI and optimization into Microsoft Dynamics 365 and the Supply Chain side most people avoid: copilots and plain-language answers over your ERP data, warehouse and routing optimization, and automated workflows that run inside the system your operations team already uses. Around 10 years building production software, the last few deep in AI, across AWS, GCP, and Azure. If it doesn't run in production, it doesn't count. WHAT I DO BEST - AI Agents in Production: LangGraph, Google ADK, AWS Bedrock AgentCore, supervisor routing, typed tool contracts, human-in-the-loop approval, and full audit trails - RAG and Document Intelligence: hybrid search, GraphRAG on Neo4j knowledge graphs, per-field confidence extraction, and citations on every answer, built for legal, fintech, and ERP data - Fixing broken AI: RAG that hallucinates, agents that fail under load, and inference bills that are too high. Evaluation harnesses and LLM-as-judge quality gates, guardrails, and token-cost engineering (I've cut client AI spend 75%) - AI and optimization inside your ERP, especially Microsoft Dynamics 365 with Azure AI and Power Apps: copilots, plain-language answers over ERP data, warehouse and route optimization, and automated workflows inside the system you already run - Document intelligence: read, classify, extract, and validate data from contracts, financial documents, and forms - AI built into a product you already use: your site, LMS, CRM, even Microsoft Word PROOF, VERIFIABLE ON THIS PROFILE An investment platform for a family office. Investment and legal documents (PPMs, SPAs, cap tables) become validated structured data, linked in a Neo4j knowledge graph. 7 AI agents answer in plain English, source attached, behind role-based access. It runs in production on Google Cloud with Vertex AI and Gemini. Contract review inside Microsoft Word. A contract intelligence platform that checks a vendor contract against your playbook, with 23 AI agents scoring every clause across 12 legal categories, flagging risky language, and suggesting fixes. A full review takes 5 to 10 minutes instead of hours, right inside Word where lawyers already work. It runs in production on AWS, and smart routing cut the AI cost 75%. A live AI co-pilot for K-12 writing teachers, embedded in their LMS. Teachers ask in plain language and get a classroom-ready lesson in seconds, grounded in their own curriculum by hybrid RAG, with an LLM-as-judge scoring every lesson on a 100-point rubric before it ships. Active client. AI and optimization inside an enterprise ERP. For an automotive-parts manufacturer with multi-region warehouses, I optimized inventory slotting and picker routes, then optimized freight routes and wired it directly into their Microsoft Dynamics 365 ERP, so it runs inside the system their operations team already uses. HOW I WORK I take the requirements, make the technical calls, and hand back working software that runs, not a list of problems. I scope before I build: architecture and acceptance criteria first, milestones tied to tests passing, weekly demos, and everything documented and handed over so your team owns it. URGENT FIXES Broken RAG or a misbehaving agent system? I take fixed-scope diagnosis-and-repair engagements: full pipeline diagnosis, root cause with evidence traces, and fixes proven against a golden set built from your real queries. Diagnosis ships in days, not weeks. TECH I WORK WITH Agents: Google ADK, AWS Bedrock AgentCore, LangGraph, LangChain, LlamaIndex, CrewAI, MCP LLMs: Claude, GPT-4o/5, Gemini, Llama, Amazon Nova, LiteLLM routing, Ollama for on-prem RAG and data: Qdrant, Pinecone, Weaviate, pgvector, Neo4j, Elasticsearch, Vertex AI Search, hybrid search, GraphRAG Cloud and ERP: AWS, GCP, Azure, Microsoft Dynamics 365, Power Apps, Docker, Kubernetes, Terraform I keep a few consultation slots open each week for architecture, cost, and feasibility reviews. If you want a senior read on your system before committing budget, book one, or invite me to your job and I'll respond within hours.