Brian M.

I build AI systems that do real work reliably: agents that use tools, pipelines that pull structured data out of messy documents, and question-answering over a company's own knowledge (RAG and knowledge graphs). What I've built: - AI "digital twins" for live customers (Steno.ai): backend services, knowledge base and retrieval for AI clones of experts and creators. I also built the voice pipeline, with blind tests of ElevenLabs, Cartesia and Voxtral voices, and the automated grading system that scores answer quality. - Company-wide AI agents (Inside Success): a personal AI agent in Slack for each member of a 19-person department, with 80+ tools across Slack, GitHub, Monday, Notion, Google Drive, Gmail and calendar. It takes per-person sign-in and asks for approval before any action that changes something. - Document-to-data pipelines: turning thousands of PDFs, transcripts and reports into structured records and knowledge graphs, with every extracted fact linked back to its exact source passage. - Research and analysis at RAND (5 years): AI and data work for DHS, the U.S. Army and RAND itself, including processing ~6 billion Reddit comments on AWS and analyzing 100M+ tweets. How I work: AI reads and drafts, plain code makes the decisions that must be exact, and every output can be traced and checked. I write tests, document what I deliver, and tell you early if something won't work. Good fits: LLM/ChatGPT/Claude integrations, AI agents and automations, PDF/email extraction into Sheets or databases, RAG/chat over your documents, web scraping and data pipelines, Python tools, evaluation of AI output quality, and research or data analysis.