Ciao!

I am Alessandro, 28 years old, from Lecce. In 2014 I moved to Milan to study “Ingegneria Informatica” at Politecnico di Milano. I got my Bachelor Degree and decided to keep on studying here, so I enrolled in the Master program “Computer Science and Engineering”, focusing on AI and Machine Learning. During the program, I also spent one semester abroad at Chalmers University of Technology, in Gothenburg, Sweden.

At the end of my studies, I had the amazing opportunity of working on a couple of research projects with the Institute of Applied Computational Science at Harvard University, and we published a paper at the NIPS 2020 Workshop Machine Learning and the Physical Sciences.

In this moment I’m working as a Data Scientist at Casavo, but you can check an up-to-date version of my resume here.

In this page I will collect some bits of my academic and job career, and whatever I’d like to share with the rest of the world.

💻 Currently working at…

Casavo is a proptech born in 2017, whose mission is to change how people sell, buy and live houses in Europe. I’m currently leading a team of 3 Data Scientists in the Pricing Tribe, and I have end-to-end ownership of strategic Data Science initiatives for all the EU countries in which Casavo operates. Following you can find a non-exhaustive list of projects I worked on at Casavo:

EVA, Casavo’s Automatic Valuation Model (python, lightGBM, docker, DVC, MLFlow, AWS)
Brunelleschi, a Language Model that extracts information from property descriptions (python, pytorch, huggingface, OpenAI, docker, DVC, MLFlow, AWS) … watch a demo here!!!
Cloyster, a Computer Vision classifier that catches buildings in bad conditions (python, pytorch, docker, DVC, MLFlow, AWS)
Doduo, a tool that finds duplicated online listings by using Network Science (python, networkX, docker, Argo Workflows)

Most relevant personal projects…

Semi-supervised NNs for COVID-19, published at NIPS 2020 Workshop - In this work, we study the spread of COVID-19 using a semi-supervised neural network and assuming a passive part of the population remains isolated from the virus dynamics. We applied a novel method to solve differential equations with Neural Networks. (python, pytorch)

From Potential To Rays - In this research project we leveraged Deep Learning to study a well-known optics problem, namely trying to predict where caustics will emerge, given the position and the intensity of the rays hitting a surface. The proposed architecture consists of two Variational Autoencoders and a Feed-Forward Network to map the two latent spaces. (python, pytorch)

NetflixGPT Recommender - This project implements a Recommender System for something to watch on Netflix based on your mood! You are asked to describe how you’re feeling today and it will recommend something to watch that gives out the same vibes. It leverages OpenAI’s LLMs to embed text and a vector database for embeddings queries.(python, langchain, llms, milvus) … watch a demo here!!!

PotterGPT - This is an educational project I built in order to learn how LLMs works. I rewrote GPT from scratch and trained it on the 1st book of the Harry Potter’s saga. (python, pytorch) … watch a demo here!!!

Neural Style Transfer - Pytorch implementation - A Pytorch implementation of the paper A Neural Algorithm of Artistic Style by Gatys et al. Neural Style Transfer is an optimization technique to manipulate an image and apply the visual appearance of another one. (python, pytorch)

(see more below…)

Skills

Machine Learning libraries

numpy, pandas, sklearn, pytorch, transformers, langchain, networkx

Data

postgreSQL, dbt, snowflake, neo4j, milvus

MLOps

MLFlow, Argo Workflows, docker, AWS

Main blogposts I published…

(see more at my Medium profile)

Previously worked at…

Moviri Analytics combines big data engineering and data science expertise to build bespoke, end-to-end business analytics solutions at scale, creating unique value for our customers through cross-industry innovation. The following is a summary of the activities I carried on during my time there as a Data Scientist.

Anomaly detection on financial time series (python, splunk)
Network analysis for incident prediction (python, networkx, neo4j)
POC to automatically translate Naturale Language into SQL query (python, pytorch, huggingface, flask, AWS)
Design of project solutions for a range of clients
In charge of the technical assessment of the recruitment process

Institute of Applied Computational Science at Harvard (IACS) is the home for students and faculty who are tackling major challenges in science and the world through the use of computational methods. I was selected to work on a combined Data Science project, delivered by Politecnico di Milano and Institute of Applied Computational Science at Harvard University, regarding advertising strategy optimization.

Consequently, we started a research collaboration to work on my Master’s Thesis, whose title is: “Improving domain-specific Transfer Learning applications in Image Recognition and Differential Equations”.

Lastly, we worked on the application of semi-supervised learning for the estimation of the basic reproduction number of COVID-19, more info in the projects section.

Our research collaboration lasted about one year after my graduation.

And other personal projects…

Improving Transfer Learning - My Master’s thesis in Computer Engineering at Politecnico di Milano, developed at Harvard University. Improving domain-specific transfer learning for image recognition and differential equations.

CartPole-DQN - This repository contains Pytorch code to train a Deep Q-Learning network that learns to balance a CartPole, a common Reinforcement Learning benchmark.

Breanna - A Python implementation of a tool to manage advertising campaign data and build Machine Learning models on them.

MSL-Datashack2019 - Breanna… explained.

Lorenzo il Magnifico - Java implementation of board Game “Lorenzo il Magnifico”.

Sales prediction for Data Mining course - A bunch of Python notebooks we used for our Data Mining and Text Mining project.

Fake News Detector - A Deep Learning pipeline to train a fake news detector

People I met…

Polimi Data Scientists is a student association I founded, with the aim of connecting the world of work with the academic one, focusing on the data science field.

Arranged 10+ talks and meet-ups with prestigious companies and professors, with 50+ attendees
Published articles and course notes on Data Science topics
Gave educational support to younger students willing to pursue a Data Science career

Notes I wrote…

“Data Mining and Text Mining” - Answers to Exam Questions

Alessandro Paticchio