Marco Pietrosanto

A system that predicts the energy that will be produced from renewable plants in the short-term, i.e., between one and 12 hours. I communicated with partners and external collaborators as the project and technical lead. Also, I directed and actively developed the ML core.

The system uses an ensemble model that directs other sub-models' contributions following a hypothesis that eolic energy production does not follow one specific regime, but each sub-model contributes dynamically throughout a certain period of time. I came up with the concept of an orchestra director for the storytelling, so the specific models were named after particular music compositions, like Minuet for the eolic energy and Bolero for its photovoltaic iteration.

A system that predicts short-term environmental risks. It integrates social posts, weather data, satellite images, and population information to train a multi-XGBoost model to predict risk values for people, buildings, and the environment.

After being put in charge of the project halfway through, i.e., in the last 1.5 years, I had to rebuild it entirely while respecting contracts and grants of around €2 million. Additionally, I had to be a scientific validator, understand the needs and capabilities of all external collaborators, and communicate with all the stakeholders to help them understand what the project was really about. Finally, I discovered that one of the collaborators did not deliver the ML core of the project even though everyone else before me thought he did, so I built it myself.

A B2B service designed for the HoReCa field provides support in analyzing and forecasting product demand. The system is able to cluster providers, analyze product time series, and provide suggestions on what products are sold together.