📍 Milan, Italy · Rust + TypeScript + Python · Building the reliability layer for AI coding agents
The models are getting commoditized. The durable value is moving to the layer that makes any agent trustworthy on real codebases: the right context going in, and verified work coming out. I'm building that layer, and it runs entirely on your machine.
Vexp · The local-first context & verification layer for AI coding agents
AI agents burn most of their budget exploring a codebase before writing a line, then "finish" tasks while leaving broken references three files away. Vexp fixes both ends. It builds a live dependency graph of your code at the AST level and feeds any agent only what the task touches (87% fewer tokens per call), then, when the agent declares it done, verifies the work mechanically against that same graph. No model grades anything; the verdict is git, tree-sitter, and the code graph.
Solo-built the full stack: a native Rust daemon and code-graph engine, an MCP server, a VS Code extension with an LSP bridge, an ambient per-prompt classifier, and a self-hosted SDK, shipping to real users across 34 languages and 14 agents.
| Agents supported | 14, including Claude Code, Cursor, Copilot, Windsurf, Zed, Codex, Cline, Aider |
| Languages parsed | 34, via tree-sitter |
| Architecture | Single native binary · 100% local · zero network calls |
| What's new (2.5) | Mechanical work verification · local PII/secret scanning |
→ vexp.dev · VS Code Extension · Open VSX · npm CLI · Horizon (open source)
Brain0 · The black box for AI-written code (open source)
A passive decision graph linking every commit to the agent prompts behind it: drift detection between what an agent declared and what it actually changed, DLP audit of what agents read, evidence-driven risk scoring, MCP memory for coding agents, signed provenance attestations. One command, offline by default. Rust core (14 crates), TypeScript/PixiJS graph GUI, Apache-2.0. 395+ stars.
Quibo · Rank everywhere. Write nothing.
Agentic SEO content engine: a 16-step multi-agent pipeline that researches, writes, and publishes on-brand articles straight to your CMS (WordPress, Sanity, Framer), optimized for Google, Bing, and AI search. Dogfooded on vexp.dev: 4.1% Google CTR and 674 AI citations in 90 days, roughly 3x the web's average organic CTR.
Lumigo · AI-powered entertainment discovery
NLP mood-based content recommendations. 500K+ page views, 30K users, €0 marketing. Owned product strategy, architecture, and go-to-market end to end.
I find a real problem, build the whole thing myself, ship it, and get users. Consistently, without budgets.
| Year | Project | What happened |
|---|---|---|
| 2026 | Vexp | The reliability layer for AI coding agents: context in, verified work out |
| 2026 | Brain0 | Open-source black box for AI-written code: provenance, drift, DLP audit (395+ stars) |
| 2026 | Quibo | Agentic SEO engine (16-step multi-agent pipeline): 4.1% Google CTR, 674 AI citations in 90 days on vexp.dev |
| 2026 | Vaultscan | Security scanner for AI-built apps: one scan, one paste-ready fix prompt (Lovable, Replit, Bolt, Cursor) |
| 2025 | Lumigo | AI entertainment platform: 500K page views, 30K users, €0 marketing |
| 2022 | Fantagoverno | Political engagement game: MVP in 4 weeks, 6K users in 3 months |
| 2021 | Clicmon | Gen Z sustainable marketplace: 9-person team, 42% waitlist conversion, €1.32 CAC |
Team Leader → Software Analyst at Intesa Sanpaolo (via Exprivia S.p.A.)
Six years at one of Europe's largest banks. Started building enterprise integrations across trading, front-office, and core banking (IBM MQ, WebSphere, SQL, Python, Unix), cutting deployment time 40% via CI/CD. Promoted to Team Leader: led a hybrid team of 6 across finance and IT, owning SLAs, budgets, and direct stakeholder relationships with trading desks. 99.9% uptime. ECB/ECMS regulatory compliance delivered on schedule.
