Matthew M.

I build production AI systems end to end: RAG chatbots, LLM apps and multi-agent workflows (LangChain, LangGraph, MCP) that make it out of the notebook and into your product. 8+ years in applied ML and deep learning, M.S. from Carnegie Mellon's School of Computer Science, based in San Francisco. RECENT BUILDS • Enterprise RAG and chat system with agentic context engineering (ACE, from Stanford's Zou Lab), with tracing, eval datasets and a RAG metrics dashboard so the team can see exactly what retrieval is doing • SFT fine-tuning of Llama models plus an LLM-as-a-Judge evaluation pipeline for MultiOn's autonomous web agent (Chrome extension and API) • Co-founded EzOut and built its YOLOv7 retail vision pipeline: labeling, training, TorchServe deployment and Prometheus/Grafana monitoring, plus the Django/React customer service • 1st place, AGI House MCP Build Day: an MCP-powered brand SEO analytics agent deployed to Fetch.ai's Agentverse and ASI:One WHAT I CAN BUILD FOR YOU • AI agents and multi-agent systems: tool/function calling, MCP servers and integrations, autonomous web agents (LangChain, LangGraph, AutoGen, OpenAI, Gemini, Claude) • RAG and LLM applications: enterprise chat, copilots, document and scientific Q&A, with vector search, hybrid retrieval, systematic evals, tracing and dashboards • Model customization: SFT fine-tuning (Llama), LLM-as-a-Judge eval pipelines, prompt engineering with monitoring • Computer vision: YOLO, Vision Transformers and CNN pipelines for detection, segmentation and health-metric extraction, deployed with TorchServe and monitored in production • Full-stack and MLOps around the model: Python, TypeScript, React/Next.js, Django, AWS, GCP, Docker, Kubernetes, Terraform, CI/CD WHY SCIENCE AND HEALTH-TECH TEAMS HIRE ME I started in the lab, not just the IDE: spatial biology, NGS, PCR and GLP lab automation at Rebus Biosystems and Prenetics, a co-authored spatial transcriptomics conference paper, blood-based organ aging clocks on NHANES data (PyTorch, Spark, Cox survival models), and digital pathology research at CMU benchmarking Vision Transformers against CNN ensembles. Your domain experts never have to translate for me. HOW I WORK You work directly with me on architecture, code, deployment, monitoring and documentation. Expect working code and a demo early (usually within the first week), clear async updates, and a handover your team can own. Open to short sprints (POCs, RAG audits, eval pipelines) and long-term builds. US-based, Pacific time, fully remote. Send me what you're trying to build and I'll reply with exactly how I'd build it: architecture, stack and timeline.