These are the machine learning, deep learning and LLM projects I built at Politecnico di Milano, written up as case studies: the problem, the approach, my own part and, where available, the measured results.
Some are individual projects, others were carried out in a team with other colleagues: each write-up describes my own contribution.
My MSc thesis, started in September 2026, studies deep learning architectural strategies for Streaming Continual Learning. It is at an early stage: working title, research question still to be defined, no results yet.
Project for the Embedded and Edge AI course: estimating, frame by frame, the position of 0–4 people in a room from the signals of 6 UWB radars, with a model that fits the memory limits of an ESP32-S3.
For the Natural Language Processing course at Politecnico di Milano I built the core of an agent that plays a timed quiz, in text and speech modes: web search, retrieval and reasoning with a local open-weights LLM.
I implemented Continuous LSTM and Continuous PNN as CapyMOA-compatible classifiers and evaluated them on the Weather stream: cPNN fluctuates less in most concepts, while cLSTM has the higher mean Kappa in 5 of 8.
Profiling, quality measurement, cleaning, imputation, record-linkage deduplication and a final re-assessment of a real-world shark attack dataset, built for the Data and Information Quality course.
Kaggle challenge of the AN2DL course: classify 160-step multivariate sequences into three pain levels, with heavily imbalanced classes. Team result: 95.5% F1 on the Kaggle leaderboard. My part: the v2 model architecture.
Working on something similar?
Tell me about it: I'll reply by email.