I design and build agentic AI systems that run in production, and I make them economically viable per transaction. I design the system, then I build it.
Cloud governance and provisioning at global-enterprise scale in a 100% infrastructure-as-code environment, and agentic AI platforms taken from prototype to production for clients.
Model-per-task selection, caching and cost-per-transaction budgeting treated as first-class architecture constraints, not an afterthought once the bill arrives.
I have trained 200+ people on enterprise cloud and collaboration platforms, and I advise non-technical directors today. I can defend an architecture to a board and to an engineer in the same week.
Systems I designed, built and operate end-to-end at Minka, my AI consultancy and R&D lab. The multi-agent orchestrator ran from May to September 2026 and is paused; the client assistants, the CRM and the release gates keep running.
An orchestrator agent delegating to seven specialist agents, with model tiering by role and a five-model fallback chain whose last hop is a different provider. Built on OpenClaw, a third-party agent gateway; paused since September 30. The client-facing assistants keep running.
Three methods compared on 6 labeled queries over 181 documents (May 2026): recall@5 of 53.7% with the multilingual model in production (55.8% with mpnet) vs 30.8% for BM25. It did not reproduce on the larger October corpus, and I published the correction. Unit suites run on every push; retrieval and routing evals run weekly in CI and on demand.
Client bots run tool-less and request capabilities through a fail-closed broker with per-tenant spend caps, a global monthly ceiling and human approval gates. Public bots cannot spend at all.
Domains, email authentication, three clouds (AWS, Railway, Vercel) and CI: 6,000+ automated assertions across 78+ suites on every push; production advances only from a green build, with deploy-drift alerts.
A self-hostable TypeScript app for 3D-printing businesses: live costing, IP and license risk scoring, and AI-generated listings. View on GitHub
A live AI web chat on a client's site that relays each conversation to the owner, database-per-client CRMs, business websites, and a payment-to-enrollment pipeline that processed real registrations for a client's workshop.
Agentic AI systems for SMBs and mid-market clients: multi-agent orchestration, tool-calling, retrieval with measured recall, conversational assistants and CRM-integrated automation. Internal capabilities exposed as an MCP server. Advised Jalisco small businesses applying to SEDECO's MiPyME digitalization program: diagnostic sessions, technical proposals and quotes under the program's rules.
Owned AWS account and infrastructure provisioning for engineering teams across a global organization. IAM policy design, permission models and security assurance for multi-team environments, in a 100% infrastructure-as-code environment (Terraform / CloudFormation).
Reference consultant for Microsoft Power Platform, project lead for a DocuSign eSignature initiative, primary LEAN champion.
Dozens of cloud migrations in the first wave of enterprise cloud adoption. Supported the IT team of the 2011 Pan American Games. Instructor-led courses on Office 365 and Azure.
An office backup product on open-source software, adopted by a team at EGADE Business School.
Spanish (native) · English (professional working proficiency). Based in Guadalajara, Mexico; remote-ready with US/EU overlap.