AvailableItaly / Remote
Software & Applied AI Engineer.
I help founders and teams turn rough ideas into working products — from architecture to production.
idea+problem
becomesunderstand
simplify
design
build
software
01
No technical spec? Fine. I help shape the scope, design the system and build the first working version.
02
Business workflows, domain-specific platforms: complex requirements turned into software that's simple to use and built to last.
03
LLMs, RAG, MCP and automation — designed as part of the product, not bolted on as a demo.
04
If a repetitive process is eating time, I figure out what to automate — with software, AI, or both.
Selected work
01
A SaaS product for reconstructing road accidents: physical models — Delta-V, deformation energy, velocity from braking traces — with results an expert can check by hand. Plus DXF comparison, expert reports and multi-user licensing.
DXF comparison
drawing A
drawing B
impact area A ↔ B
Δ length · Δ width · Δ depth
Pick two DXF drawings, select the impact area on each, compare them dimensionally.
02
A SaaS platform for customers, payments and commercial processes, with AI designed in, not bolted on: people ask in plain language and the assistant answers from the company's documents and business data.
How did payments go this quarter?
reading the uploaded documents and the business data…
An answer written from the retrieved documents and business data.
The assistant decides whether a chart helps before building one.
03
A platform for public-sector organizations running tender procedures: deadlines, shared documents, roles and permissions — with assistants that answer questions over regulations and internal policy, cite their sources, and generate the resulting documents.
Which deadlines apply to this procedure?
An answer assembled from the retrieved documents.
Every answer points back at the document it came from.
04
An open-source MCP server in Python and FastMCP that lets any compatible LLM client talk to a Trakt account: ask what you watched, and the model reaches the real API instead of guessing.
what did I watch last week?
calling trakt.history…
built mostly because I wanted it to exist.
Also built
Booking and certification workflows for a public health service, with the records and documents behind them.
Internal workflows, forms and operational tooling used day to day by a financial-services operation.
Pilots submit an intended landing and location; managers check capacity and accept or reject the request.
Structures, areas, burial locations and records, with container types, PDFs, receipts, invoices and payments.
Assets, work orders and maintenance workflows, plus the APIs and the interfaces the field teams actually use.
How I work
The cheapest place to fix a bad idea is the conversation before it becomes code. I want to know why before I agree on how.
No abstraction, service or piece of infrastructure gets in without a reason. Simple is not careless: quality still matters, it just stops growing on its own.
A four-week MVP and a system meant to run for years are not the same job. Matching the engineering to the actual lifespan is most of the work.
Requested solution
The actual problem
What's the goal?
Need to learn?
Prototype
optimize for learning
Need to last?
Production
engineer for longevity
Ship
Learn / iterate
Same care either way. Very different amount of machinery.
About
I'm Fabrizio. Seven-plus years of professional software development, several as an employee, before going freelance for what I was looking for: freedom over what I work on, and how.
What I enjoy most is problem-solving — software architecture above all. I like deciding what a system should be before deciding how to write it, because the choices made there shape everything that follows.
And I still like the simple fact that, at the end of it, you've built something that wasn't there before.
Contact
Early idea? Existing product? Weird problem? All good.