I use agents and automation as a multiplier: from prototype to the server running it, I get there on my own.
I build complete products: sites, apps, automation, middleware, and the infrastructure they run on. Agent security is one of the areas I go deepest on, and AI Guardian Lab below is the clearest proof of that.
400-hour Master in AI Development, ProfessionAI, 28/30, covering LLMs, agentic AI, REST APIs and MLOps.
Based in Turin, Italy. Open to remote roles.
Security middleware for AI agents. Shell commands pass through an allowlist, a regex layer, and an LLM intent check before running. Every decision is logged in a hash chain signed with HMAC-SHA256, so a tampered entry breaks the chain.
Input ──► L1: Allowlist ──► L2: Pattern Engine ──► L3: LLM Intent Check ──► Agent
│ │ │
BLOCK BLOCK BLOCK + log
Self-hosted task manager for two people I actually live with. FastAPI backend, React frontend with no build step, runs in Docker.
Offline-first PWA for shared-household cleaning tasks. Data stays on the paired phones and syncs automatically when they are on the same Wi-Fi.
Hand-built static portfolio in vanilla HTML, CSS, and JavaScript, deployed on Cloudflare Pages with WCAG AA accessibility as a design constraint.
Gross-to-net salary calculator for Italy, built for a hiring test. Vanilla JS, no backend, every INPS/IRPEF formula visible in the code. Live demo.
PWA for MotoGP 2026 standings and a private Fantasy league with friends. Vanilla JS, no build step, installable and works offline.
AI & Security Python LangChain scikit-learn Pandas
Backend & Infrastructure FastAPI Django Docker
- Agent safety, the focus of AI Guardian Lab above.
- Workflow automation with n8n and Docker, some of it running on my own server.
Cycling mountain routes in the Alps, and following motorsport data closely (BoxBox above started as an excuse to look at it more). Hands-on training and personal practice in 2D/3D CAD, 3D printing, and laser engraving/cutting. Projects on Instagram: @luke_lab_3d.