I build AI-powered software for cybersecurity at Drivesec. The job splits fairly evenly between writing code and
writing the functional and technical analyses that turn an idea into something a team can actually implement: I
like both halves, and I think the second one is what decides whether most projects succeed or fail.
Outside work I'm an avid reader: science fiction and crime novels fill my evenings. My interests range across
technology of every kind: from major engineering works to residential home-automation systems, from Formula 1
innovations to their practical applications, from the largest datacentres to the newest smartphone.
Multi-agent simulator in Rust · 2026 · repository
A from-scratch rewrite of the thesis work below, on real geography, built to answer the cost
question the earlier version could not. The answer is yes, and the interesting part is what it
costs: thirteen enrolled drivers serve 85% of thirty-two riders at 6.8 vehicle-kilometres per
passenger carried, pushing the fleet to thirty-two vehicles gains another ten points of service
rate and costs 13.9, and pushing it to a hundred and thirty-two adds nothing and costs 58.7.
Seats are worth five times what vehicles are, the ceiling is walking distance rather than
matching, and six one-way lines drawn by hand beat every network derived automatically from the
same stations, because they know the direction of the demand. A language model plugs into the
same dispatcher seam the fixed rules use and is measured on the same columns, with cost per
decision beside them.
First author · PFIA 2025, Dijon, France · pp. 86–94 · July 2025
A multi-agent simulation of autonomous on-demand mobility, asking whether the network a service runs on should be
structured or left free-form. Structured wins, and not marginally: more efficient routes, fewer kilometres
driven, and a service that behaves more like collective transport than a fleet of taxis. The same structuring
works for carpooling in peri-urban areas: an on-demand service built around a few trunk lines, which we called
Mobility as a Network.
Ridesharing: an analysis of performance in low-flow networks
MSc thesis · Politecnico di Torino · 2025
Three different techniques aimed at one question: can a ridesharing service survive on a network carrying very
little traffic? The honest answer was no, or at least it was not possible to prove otherwise. None of the
behavioural levers I analysed moved the number of participants enough to turn an unstable network into a working
one. Anticipated departure declaration showed a path only under forced 100% adoption, and even there the
simulation data yielded no clear statistical separation between scenarios. What did come out were hard cost
figures for low-flow operation: enough to make a strategic decision with at company level, but not a solution.
Negative results are still results.
