Usman M.

🧠 RAG & MCP-powered AI systems. Document intelligence for regulated industries. Answers you can trust and trace to a source. I build RAG systems and document-intelligence tools for businesses that can't afford their AI to guess, healthcare providers, legal teams, and financial and compliance-focused companies working with sensitive, regulated documents. 💼 Why Work With Me I have three years of focused experience in AI and data science and in software development, including published research on abstractive summarization for low-resource languages (Urdu), work that shaped how I think about grounding, evaluation, and reliability in NLP systems, not just getting a model to run. Recent work includes a compliance-verification RAG agent for finance, healthcare, and legal teams, a document redaction tool that sanitizes PII across legal, healthcare, and financial records with a full audit trail, and rebuilding an underperforming client RAG pipeline with DSPy, replacing manual prompt tweaking with a measurable, compiled retrieval framework. 🛠 What I Do RAG pipelines, built and optimized, including auditing and fixing underperforming retrieval systems, not just building new ones from scratch. Document intelligence: extraction, redaction, and compliance-document Q&A with citations, for teams handling sensitive records. Grounded LLM chatbots, MCP-based agent workflows, and automation: internal knowledge assistants, meeting-to-action workflows, and reporting tools that connect to the tools your team already uses. 🏥 Where I Help Healthcare, legal, and financial-services teams handling sensitive, regulated documents: RAG-based compliance Q&A, PII redaction, and internal knowledge assistants that cite their sources instead of guessing. ⚙️ How I Work I start by understanding your data, documents, and where the current process breaks down. If you already have an AI or RAG system in place, I'll tell you honestly what's working and what isn't, rather than rebuilding everything from scratch. Every pipeline I build is measurable, auditable, and built to hold up under real use, not just a demo. 🔧 Tools LangChain, DSPy, MCP, Pinecone, Chroma, pgvector, Hugging Face Transformers, Python, FastAPI, PostgreSQL. If you're working with sensitive documents, need answers your team can trust and trace to a source, or have a RAG system underperforming, let's talk.