What Giacomo Beccani works on

Agents will do more and more of the work. I make sure a person stays in charge: every action approvable, logged and, where it matters, signed.

• Agents take the repetitive work: posts, reports, finding contacts, looking after websites
• You approve before anything goes out, even straight from the chat
• Every action has a name and a time: you always know who did what
• Risky actions need a signature, and mistakes can be undone

Every company thinks its problem is unique. Most are not. I start from how you work, find the part that repeats and make it reusable, so the next step costs less than the first.

• I start from how you work today, not from a generic template
• I find the steps that repeat and automate them
• Custom work, built from pieces already proven in production
• You see the software working before you decide

Soon people will meet a company through an AI answer before they ever see its website. I build sites for readers, search engines and AI assistants, so the people looking for you find you.

• A website your customers find on Google and in AI answers
• Numbers instead of opinions: performance, accessibility and SEO measured before going live
• Several languages without duplicate pages to maintain
• The site stays yours: a standard theme, transferable, no lock-in

The best software disappears into the hands of the people who use it. I build native apps, signed and published, and the firmware of the objects they control, keeping data on the device whenever I can.

• A native app, signed and ready to ship
• Data stays on the device whenever it can, not on someone else's servers
• End-to-end encryption where privacy really matters
• The firmware too, if the app controls an object

Software is worth as much as the trust you can put in it. I build it solid and secure, and I design it to respect the rules that apply to you, from privacy to compliance.

• Security designed in from the start, not bolted on at the end
• Visitors' consent and preferences recorded and provable
• E-invoices checked before they are sent
• Accessibility and best practices verified, not promised

I train models for the most varied tasks, from a few million parameters to trillions. And I build the memory AI needs to know what is true, and since when.

• A model trained on your task, not a generic one bent to fit
• The right size for the problem: small and local, or large and distributed
• Memory that recalls what you decided and when, without losing the history
• Results measured and published together with their limits