I'm Dani, a senior software engineer in Madrid. By day I work on open banking at Tink (Visa); I'm also studying MLOps at Universidad Pontificia Comillas (ICAI).
What I build in the open revolves around one problem: giving AI agents the right context, and proving that what they produce actually works. That means persistent memory for codebases, MCP servers that connect models to real tools, and ML research that is reproducible before it is impressive.
- 🧠 Agents & context engineering: memory, retrieval and tool access (MCP) for coding agents
- 📊 Rigorous ML: walk-forward evaluation, probability calibration and controls against backtest overfitting
- 🏦 Production engineering: APIs, reliability and security in a regulated fintech environment
📫 Open to conversations about AI engineering, agent tooling and ML platforms: message me on LinkedIn.
| Context first | Understand the system, its constraints and the real goal before writing code. |
| Evidence over vibes | Tests, evaluations and reproducible runs decide whether something works, and "no edge found" is a valid result. |
| Automate the boring parts | Scripts, CI and agents where they remove repetition, not judgment. |