Director of Engineering at @matecat and Matesub, part of @translated. Rome, Italy.
Most of my work is translation technology — CAT tooling, subtitling, the infrastructure underneath both — and lately the tooling that lets AI agents work on real codebases without making things worse.
- Translation platforms — MateCat, an open-source AI-driven translation tool for language professionals, and Matesub, a professional AI-driven subtitling editor.
- Agent tooling — MCP servers, skills, and quality gates that give coding agents a deterministic surface to work against instead of guesswork.
- Databases and infrastructure — drivers, replication setups, and the Docker/deployment plumbing that keeps the above running.
Pinned on my profile, plus a couple more:
| Project | What it is |
|---|---|
| matecat/MateCat | The MateCat translation tool — open source, AI-driven, built for language professionals. |
| Signet-AI/signetai | Persistent memory and identity for AI agents, shared across harnesses instead of trapped in one session. |
| matecat/xliff-parser | XLIFF parser in PHP — the format every CAT tool has to agree on and none agree on completely. |
| bruno-mcp-studio | MCP server for Bruno. An agent authors, reads, and runs .bru collections with byte-parity to Bruno itself — no bru CLI required. |
| Skills | Agent skills: adversarial plan review, parallel planning, QA plan generation, teaching a change until it is understood. |
| test-guard | PR gate for test adequacy — coverage numbers plus heuristics plus a model, rather than a coverage threshold alone. |
| signet-first | Memory-first protocol for agents: search before acting, store the conclusion, hand off between sessions. |
An agent reviewing its own work rationalises. Several independent ones, given hostile framing and no sight of each other's output, do not. Most of what I write in this space is some version of that idea.
- LinkedIn — domenicolupinetti
- MateCat — matecat.com