I design conversational AI agents for companies and put them into production. Most of them sell or quote over messaging and voice, connected to the client's ERP and CRM. I work in Python with LangGraph, Gemini and Google Cloud Run. Before an agent ships, I evaluate it against golden sets of real business scenarios.
My rule for these systems is that the model handles the conversation and deterministic code handles anything with a correct answer: prices, stock, terms and routing.
Industrial engineer and MSc in Artificial Intelligence, Pontificia Universidad Católica de Chile. Before AI agents, I spent almost six years in analytics and BI at a Fortune 500 financial services company. Based in Santiago, Chile.
- voice-quote-agent: a prototype voice agent on ElevenLabs Agents that quotes container rentals. The quotes come from webhook tools, and negotiation goes to a person.
- agent-golden-set-eval: golden-set evaluation for a quoting agent. It uses deterministic checks, and CI blocks the release if a single quote is wrong.