Passionate about AI, Deep Learning, and Data Science, I specialize in developing cutting-edge machine learning models for real-world applications. My expertise spans Computer Vision, Probabilistic Modeling, and Large Language Models, with hands-on experience in research and industry.
With a Master’s degree in Data Science and a Bachelor's degree in Statistics from La Sapienza – University of Rome and an Exchange semester at the University of Helsinki, my academic background is rooted in statistics, probabilistic modeling, and AI.
Beyond my professional work, I am passionate about bridging AI research and industry applications, constantly seeking innovative ways to leverage machine learning for impactful solutions.
Work Experience
Deep learning scientist - intern
2024 - 2025
Zurich, Switzerland
I developed Computer Vision models for the improvement of the Quality Control of the precipitation measurements in Switzerland. The project involves the use of Python’s Deep Learning framework (PyTorch), Data Visualization, and Big Data Management.
Tasks
— Literature review on precipitation modeling with deep learning models
— Development of an innovative CNP (Conditional Neural Process) model for gridding point measurements on the ground, which describes both the presence and the amount of precipitation (gamma hurdle model)
— Development of a U-Net model for combining gridded ground measurements with radar fields and other explanatory variables (such as topography)
— Implementation and training of the models in PyTorch
— Qualitative and quantitative evaluation of the models and comparison with methods currently used by MeteoSwiss to combine ground and radar measurements
— Documentation of the results achieved and presentation in several internal talks for both management and specialists
msc thesis in deep learning
2023 - 2024
Zurich, Switzerland
I've written my MSc thesis at Eawag on the mathematical dynamics within neural networks at initialization. I then worked as an external collaborator for the publication of the project. The title of the thesis is “Normalization Effects on Initial Guessing Bias”, and it will soon be submitted to an AI Conference.
Thesis overview
→ Intuitively an untrained neural network tends to assign proportionally every sample to the different classes (e.g. 50% in binary classification)
→ Actually some architectures tend to be biased before training → we call this phenomenon "Initial Guessing Bias" (IGB)
→ We want to theoretically understand the problem of the Initial Guessing Bias
→ The objective of the thesis is to study what do BatchNorm and LayerNorm cause to IGB both from a theoretical and practical point of view
→ The thesis outcome is the following:
— BatchNorm before activation function: causes a non-amplified IGB that is caused by the ReLU activation
— BatchNorm after activation function: the IGB effect caused by the ReLU is killed by BatchNorm
— LayerNorm before activation function: IGB is amplified with the increase of the number of layers added in the network
— LayerNorm after activation function: the IGB effect caused by the ReLU is killed by LayerNorm
→ The IGB at initialization may influence also the training phase, leading the highly-biased networks to slower dynamics, and viceversa!
Junior data scientist
2023 - 2023
Rome, Italy
During my work at Hyntelo, i worked in two different departments: R&D and Data Analytics Consulting. My duties ranged from ML models creation to dashboard implementation and more and more and more... Thanks to the courses offered by the company, I also learned design the design framework with Figma.
As R&D Data Scientist, I worked with Machine Learning models, Large Language Models, Computer Vision models, Web Scraping algorithms and Vector Databases in order to improve the main product of the company (Lyriko).
As Data Analytics Consultant, I have developed Dashboards with Tableau for a leading financial company and implemented Data Driven Solutions for pharmaceutical companies.
Main Duties
R&D Data Scientist
— Implementation of Machine Learning models: LLM, Computer Vision, Vector Databases
— Implementation of Web Scraping algorithms to extract insights from websites
Data Analytics Consultant
— Development of interactive Dashboards with Tableau for a leading financial company
— Implementation of Data Driven Solutions in order to shift from a sales-driven approach to a data-driven approach for a pharmaceutical company
— Development of an interactive Dashboard with QlikSense
Education
My academic background is rooted in statistics, probabilistic modeling, Machine Learning and a lot of data-related stuff.
ALL MSc COURSES
Master's degree in data science
2021 - 2024
Master's degree focused on Machine Learning, Inferential Statistics, Data Mining, Bayesian Inference, Data Management, Networking for Big Data, Data Driven Economics, Deep Learning.
Exchange semester in data science
2022 - 2023
Helsinki, Finland
Exchange semester sponsored by the Erasmus+ program. During this semester, I covered the following topics: Intro to AI, Trustworthy Machine Learning, Cryptography, Pyspark, Statistical Analysis of Environment.
Bachelor's degree in statistics
2018 - 2021
Bachelor's degree focused on theoretical statistics, more specifically: Probability, Inferential Statistics, Multivariate Statistics, Stochastic Processes, Time Series, Calculus, Machine Learning.
University projects
—Ratatouille Model Merging to distinguish the different colorectal tissues and diseases: advanced Transfer Learning project with the purpose of obtaining the best model by mixing three twice-fine-tuned models.
—Saliency Maps of DenseNet-121 for leaf classification: implementation and training of a DenseNet-121 model to classify 38 different kind of leaves and further observe the gradients from the images.
—UK inflation forecasting with NLP techniques and LLMs: application of Natural Language Processing (NLP) techniques and LLMs to predict inflation trends in the UK using social media data, particularly tweets.
—Bayesian Analysis of the Danger of Asteroids: analysis of the hazard of 27'000 asteroids fluctuating around the Earth with Bayesian and Frequentist models.
—Art Current Classification: classification of artworks belonging to 7 different art movements using traditional Machine Learning and Deep Learning techniques.
—Stack Overflow Analysis with Graphs and Visualization: system that provides users with information about StackOverflow divided in two parts: backend and frontend.
—Web scraping of Anime reviews and sentiment analysis: implementation of a search engine over the "Top Anime Series" from the list of MyAnimeList.net with different metrics.
—Analysis of dugong's lengths using MCMC: analysis of dugongs length and application of Markov Chain Theorems.
tools
Python
PyTorch
MySQL
RStudio
Tableau
PowerPoint
TensorFlow
Figma
Machine Learning
Data Analysis
Computer Vision
Web Scraping
Dashboarding
Visual Design
LLMs
Statistics
Languages
English
Italian
German
