Selected work
These are representative examples of the kind of work I do. Some details are intentionally high-level when they involve private systems, customers or internal codebases.
GoKoan: learning science, planning and AI tutors
Context. GoKoan helps students prepare for competitive exams by combining planning, learning science and digital study materials.
My role. Co-founder and CTO: product architecture, technology strategy, engineering execution and AI capabilities.
What mattered.
- Turning available study time into personalized daily planning.
- Applying AI to educational content: questions, flashcards, true/false activities, fill-in-the-blank exercises, summaries and tutoring.
- Keeping the product useful for real learners, not just technically impressive.
- Moving from early model experimentation to production workflows.
Why it represents me. This is the clearest example of my long-term pattern: use technology to make complex learning more practical, adaptive and humane.
Related reading:
Koanly: AI content production for education
Context. Koanly is a GoKoan product for AI-powered educational content production.
My role. Technical and product leadership around the generative AI workflows behind the product.
What mattered.
- Converting content workflows into AI-assisted pipelines.
- Balancing speed, quality, review and cost.
- Designing systems where AI helps subject-matter work without hiding responsibility.
- Making content generation repeatable instead of artisanal.
Why it represents me. I am interested in AI as an operating model, not only as a feature. Koanly is the kind of product where process design, domain knowledge and model behavior matter together.
Agentic engineering: mining PRs into operational guidance
Context. Coding agents can produce a lot of code, but they fail when they do not understand the local taste, conventions and recurring review patterns of a repo.
My role. Designed and ran an experiment mining hundreds of PRs and review comments to extract recurring engineering feedback, then converted that into backend guidance, checklists and agent handoff rules.
What mattered.
- Treating PR comments as a corpus of tacit engineering knowledge.
- Separating generic rules from repo-specific conventions.
- Building a review checklist for both humans and coding agents.
- Closing the loop so future PR review comments can become new guidance.
Why it represents me. This is close to my current thesis: strong AI-assisted engineering is not “more prompting”; it is building a system where agents inherit the operating culture of the team.
Related reading:
- Lo que las PRs saben y el código no cuenta
- Las 7 leyes del desarrollo en un mundo agéntico - Parte 1
- Las 7 leyes del desarrollo en un mundo agéntico - Parte 2
Technical feedback loops with Codex
Context. A serious technical problem is not only a bug; it is a chance to improve how the team detects, reproduces, tests and prevents similar issues.
My role. Built a closed loop for detection, triage, tests, fixes and documentation with Codex as an engineering operator.
What mattered.
- Making reproduction explicit.
- Adding tests that captured the failure mode.
- Turning a one-off fix into a reusable loop.
- Using the agent for disciplined execution, not blind code generation.
Why it represents me. I care about AI because it can increase engineering leverage, but only when paired with evidence, tests and operational discipline.
Related reading:
Enterprise architecture before the AI wave
Before GoKoan, I spent many years building and leading complex software projects in enterprise environments: Ford, Tissat, Indra and Akamon.
Selected themes:
- Modernizing healthcare systems without breaking production service.
- Building search and dynamic forms for clinical workflows.
- Leading education administration systems with large multi-site teams.
- Creating reusable Java frameworks, web architecture and code-generation tools.
- Working as developer, architect, project director, product director and technical lead.
Why it still matters:
- AI does not remove the need for architecture, reliability, observability and clear ownership.
- My current AI work sits on top of a long background in systems that had to survive real users, budgets, deadlines and legacy constraints.
Related reading:
Personal AI/product lab
I also build small products and experiments to explore how AI changes learning, writing, reading and personal systems.
Examples:
- Biblioteca: lectura activa para la era de los copilotos
- MyJournalGenerator
- DreamLoop
- RunLog Pro
- Cambio de paradigma en la programación
Why it matters:
- I learn best by building.
- Small apps are a good way to test product instincts, interfaces, AI workflows and personal operating systems before they become company-scale patterns.