Backends, data pipelines, the models in the middle and the interface on top. Usually I'm the person who takes something from "we want this" to something running that people use every day.
Systems where the AI is one component among many, not the whole product. Most of the work is everything around it.
End to end: working out the actual problem, designing it, writing it, shipping it, fixing it when it breaks.
Turning a vague requirement from someone non-technical into something that works and keeps working.
I've worked as the only engineer on a product and as the person coordinating scope with clients. Both taught me the same thing: most systems fail on the misunderstood requirement, not the missing library.
Two habits I keep. Show the working, because an answer with no visible path to it doesn't get trusted and shouldn't be. And learn the user's vocabulary, because every organisation uses the same words to mean different things and the software has to pick that up rather than assume.
Degree in IoT, big data and machine learning, University of Udine. Based in Italy, working in Italian, English, Ukrainian and Russian. Currently AI Engineer and Project Manager at Tecnovation.
A multi-agent system that turns questions from non-technical people into verified queries across operational databases, then returns tables, charts and the query it ran. Handles ambiguous terminology and remembers what particular words mean to a particular user. Built and deployed solo.
Python · FastAPI · PostgreSQL · MariaDB · Redis · ChromaDB · React · Docker
Incoming supplier documents read automatically: lines extracted, matched against existing records, written into the system, and anything that doesn't reconcile flagged rather than guessed at. Replaced manual re-typing and the errors that only surfaced weeks later.
Python · document extraction · ERP integration
A chat assistant answering technical product questions for buyers, and a planning tool modelling the economics of a season's work. Different audiences, same problem: putting expert knowledge where the decision gets made.
Python · LLM agents · React
Open-source terminal tooling for monitoring distributed task queues, built because the existing options showed everything except what I needed while something was on fire.
Python · Textual · Celery
Machine learning models predicting how traffic redistributes across European airspace after a closure, trained on historical flight data. BSc thesis, 2026.
Python · scikit-learn · time series
Two ways to extend an agent that look interchangeable in the docs and behave nothing alike in production.
The rules that live in people's heads and not in the schema, and four ways to get them into a system.
Open to consulting and contract work.