At Consultorsalud, I decided the data and AI architecture using four DAR analyses, selecting self-hosted PostgreSQL, a Kimball raw-to-core-to-mart model, and a hexagonal backend. I integrated 321,850 records from official sources with 98% coverage, using HL7 FHIR R4 rules and traceable human review.
At Consultorsalud, I designed an AI layer where the model reads only curated views and an algorithm and editor determine criticality. I used RAG with pgvector, n8n, and MCP, tracking cost per conversation.
At HUB Medrar, I coordinated up to 10 squads across six SaaS product lines in one roadmap, reducing time to market by 35%. I was Product Owner of the CORE system for Titularizadora Colombiana from 2022 to 2024, and built TRIO, a multi-tenant CRM with AI agents on WhatsApp, and AXORA, an EdTech product with LLMs and RAG.
On my autonomous agent squad project, now in production, a Telegram coordinator and development and research agents deliver work through pull requests. I set up least-privilege permissions, a human approval gate, and independent auditing; across 12 rounds, eight were rejected. In a Jelou technical challenge, I made sales confirmation a domain-controlled action outside the model's reach and tested the WhatsApp ticketing agent with 80 tests and 110 criteria, with zero critical issues.

