Guglielmo P.

I build AI agents, LLM and RAG chatbots and automations that run in production. I am first author of a paper accepted at the NeurIPS 2026 HAIC workshop, drawn from the live AI platform I designed and still run, and I founded AIGROW, which is part of NVIDIA Inception and Microsoft for Startups. The work I take on is the step after the demo: connecting the model to your real data and to the tools your team already uses, and keeping it working once real users arrive. WHAT I BUILD • AI agents that do the work: multi-step reasoning, tool calling, MCP connectors, and the guardrails that stop them going off-script or running up your bill. • Chatbots that answer from your own documents, attach the source to every answer, and say so when the documents do not cover the question. • Workflow automation that takes over the task nobody on your team wants to do twice. • AI built into the app you already have, plus the evaluation that shows where a model can be trusted and where it still needs a person. PROOF YOU CAN CHECK • 1o1, a nonprofit building a dialectic AI platform, picked my proposal out of 44. I shipped a working MVP in ten days and have built and run the production system since: memory across sessions, retrieval over each user's own documents, safety checks before any model call, about 370 automated tests. Their review: "one of the most substantive technical collaborations of my career so far." • The paper that came out of that work, "Instrument before you evaluate", shows that whether an AI system can be evaluated is decided by what it records, and I design what my systems record with that in mind. • AIGROW publishes the AI Visibility Index, a monthly open dataset of which businesses ChatGPT, Claude, Gemini and Grok recommend: 80,059 answers from 20 countries in 11 languages, with a DOI. Claude and ChatGPT can query its AIGROW Score through our MCP connector, and AIGROW Labs benchmarked 36 models on 20 real office tasks. • Certified by MIT Professional Education (Large Language Models), the NVIDIA Deep Learning Institute (Building Agentic AI Applications with LLMs, scored 100/100) and Anthropic (Building with the Claude API). HOW I WORK You get a written plan before I touch code, and changes go to staging before production. My updates carry numbers you can check, I report my own bugs as soon as I find them, and if a feature is not worth building I say so. Every decision stays written down, so your team can read it later. Stack: Python, TypeScript, FastAPI, Next.js, React, PostgreSQL with pgvector, Docker, MCP, Claude API, OpenRouter, Stripe. Tell me the outcome you need and what is in the way, and you will get a scoped plan and a quote back, usually the same day.