This profile is my personal applied AI lab. I design, build, and test agents, automations, and local-first tools that turn complex workflows into useful, verifiable, and maintainable systems.
I am especially interested in the work that begins after a prototype: evaluation, traceability, security boundaries, failure recovery, artifact quality, and a clear experience for the person using the system.
- AI agents with well-defined responsibilities and outputs.
- Quality, fidelity, and evidence checks before accepting a result.
- Local-first automation with privacy, human control, and reproducible execution.
- Document, research, analysis, and assisted decision-making workflows.
- Interfaces that make long-running or technically complex processes observable.
- VerbaLoom — the flagship project of this lab: local-first AI translation for long books, with editorial memory, fidelity audits, resumable quality gates, and human review.
- Red Privada — an AI-assisted research lab for provenance-aware knowledge graphs and human review, named in homage to journalist Manuel Buendía and his column Red Privada.
- APK DocForge — a local pipeline for analyzing authorized APKs and producing evidence-backed documentation and audits.
- Market Radar — an explainable-signals lab for market data quality and decision tracking.
- Q-learning Snake — a compact tabular reinforcement-learning experiment with Python visualization and a browser demo.
- Neon Snake — a compact web game used to exercise deterministic logic, automated tests, and continuous delivery.
- Evidence matters more than a claim of success.
- Sensitive data and credentials do not belong in a repository.
- A useful workflow should be resumable, auditable, and testable.
- AI proposes or transforms; critical boundaries remain deterministic.
- Documentation is part of the product.
I am open to opportunities where applied AI, automation, and reliable systems design can solve real problems end to end.