Resume

Data and AI Reseacher

Engineering, Bari, Italy

Engage in Research & Innovation projects in diverse national and international contexts, driving solutions that address real-world challenges. Study, analyze and apply cutting-edge methods, methodologies, techniques and tools to meet project requirements and objectives. Design, develop, and test prototypes leveraging Machine Learning (ML), Advanced Analytics, and Large Language Models (LLMs) to deliver impactful, scalable solutions.

• Python
• LLM
• NLP
• Data Science
• Artificial Intelligence
• Research

Research Intern

SisInfLab, Polytechnic of Bari, Bari, Italy

Supervised by Professor Vito Walter Anelli and Professor Tommaso Di Noia. Continued the work previously done as visiting researcher at the University of Glasgow to carry out the master’s thesis. Designed and implemented S6Rec, a structured state space model (SSM) based on Mamba for sequential recommendation.

• Python
• PyTorch
• Hugging Face Transformers
• Git
• Academic Research
• Recommender Systems
• State Space Models

Visiting Researcher

IR Group, School of Computing Science, University of Glasgow, UK

Supervised by Professor Craig Macdonald and PhD student Aleksandr V. Petrov. Delveded into the world of recommender systems with a particular focus on sequential recommendations in the music domain. Augmented Transformer‑based models with personalized popularity awareness, yielding improvements which range from 25.2% to 69.8%. Submitted a paper accepted for publication in the RecSys 2024 LBR track.

• Python
• PyTorch
• Hugging Face Transformers
• Git
• Academic Research
• Recommender Systems
• Transformer-Based Models

M.Sc. in Computer Science Engineering

Polytechnic of Bari, Bari, Italy

Artificial Intelligence and Data Science

• Fundamentals of Machine Learning
• Agent-Based Artificial Intelligence
• Mathematical and Statistical Methods for Artificial Intelligence
• Deep Learning
• Information Retrieval and Personalization
• Information System Design and Big Data
• Formal Languages and Compilers

B.Sc. in Computer Science and Automation Engineering

Polytechnic of Bari, Bari, Italy

Computer Systems and Applications

• Algorithms and Data Structures in Java
• Mathematical Analysis
• Computer Science for Engineering
• Databases and Information Systems
• Numerical Calculation
• Operating Systems
• Software Engineering and Web Foundations

Scholarship for thesis abroad

Politecnico di Bari

Based on the weighted average of career exams and the qualitative characteristics of the thesis programme.

Study award for deserving high school students

High School "Galileo Galilei", Bitonto

Based on average votes, grade in the state exams and student personality.

Valorizzazione delle eccellenze

Ministero dell'Istruzione, dell'Università e della Ricerca (MIUR)

Award for achieving High School Diploma with a grade of 100/100 cum laude.

Study award for 1st place in the local competition of ”Olimpiadi di Fisica”

High School "Galileo Galilei", Bitonto

Organized by AIF (Italian Association for the teaching of Physics).

DIVAN: Deep-Interest Virality-Aware Network to Exploit Temporal Dynamics in News Recommendation

ACM RecSys Challenge 2024 (RecSys Challenge '24)

This study presents DIVAN (Deep-Interest Virality-Aware Network), our solution for the RecSys 2024 Challenge, combining a Deep Interest Network (DIN) for personalized user interest representation with a Virality-aware Click Predictor that utilizes temporal features to estimate click probability based on news popularity.

Enhancing Sequential Music Recommendation with Personalized Popularity Awareness

18th ACM Conference on Recommender Systems (RecSys '24)

This paper introduce a novel approach that incorporates personalized popularity information into sequential recommendation. By combining user-item popularity scores with model-generated scores, our method effectively balances the exploration of new music with the satisfaction of user preferences.

SciQA - a Scientific Question Answering System with Citations

Designed and implemented an end-to-end solution to provide accurate, contextually relevant, and citation-supported answers to user queries.

• Python
• Hugging Face Transformers
• LLM
• NLP
• RAG
• Prompt Engineering

DistilBERT-based Architecture for Sentiment Analysis

Designed and implemented a binary sentiment classification architecture for hotel reviews: it includes the pre‑trained DistilBERT base uncased model, a dropout layer and a fully connected layer, fine‑tuned using a large dataset from Tripadvisor, in order to accomplish the task. Obtained an accuracy score of 98%.

• Python
• PyTorch
• Hugging Face Transformers
• pandas
• matplotlib

Shipping Delays Analysis and Prediction

Performed exploratory data analysis to understand the factors influencing shipping delays and identified clusters that bring the highest revenues to the company. Developed predictive models (Linear Regression and XGBoost) to forecast shipping delays and quantify the expected delay in days. Tailored the training and evaluation process by giving more weight to the most profitable customers.

• Python
• PySpark
• SQL
• pandas
• Power BI
• Data Science

word2owl – Mapping Natural Language to OWL and Back

Designed and implemented a system, based on LLM and RAG, able to convert natural language sentences into OWL 2 declarations and axioms (insertion mode) and to answer queries in natural language concerning its memorized knowledge (query mode). Designed a RAG technique, based on a knowledge graph and a vector index (faiss with embedding based on a SBERT model), to include in the system prompt the available knowledge related to the user’s query. Designed 4 different prompt strategies for the system’s insertion mode: few‑shot, few‑shot Chain‑of‑Thought, prompt chaining few‑shot, prompt chaining few‑shot Chain‑of‑Thought. Defined evaluation measures and created an ontology to manually test the system, both in the insertion mode and in the query mode.

• Python
• Ollama
• Git
• LLM
• RAG
• Prompt Engineering

spamBERT – Spam Text Classification Using BERT

Designed and implemented a spam classification architecture for email and sms texts: it includes the pre‑trained BERT base cased model (with 12 encoders) and a fully connected layer, fine‑tuned using SMS Spam Collection and Spam‑Ham Dataset, in order to accomplish the task. Implemented a simple user‑friendly webpage for model inference. Obtained an accuracy score of 99%.

• Python
• PyTorch
• Hugging Face Transformers
• scikit-learn
• pandas
• Git
• NLP
• Fine-Tuning

outGANfit – Compatible Outfit Generation Using GANs

Designed and implemented a Conditional Deep Convolution Generative Adversarial Networks (cDCGANs‑)based architecture, trained with images from the Polyvore Outfits dataset, to generate outfit items compatible with the input garment: it takes as input a t‑shirt and generates matching pant, pair of shoes and accessory. Included 3 different GANs, one for each garment to be predicted, composed by 1 generator and 2 discriminators: the 1st used to evaluate the compatibility between the conditioning image and the generated garment; the 2nd used to classify what is generated as real or fake. Obtained a FID value of 175.42.

• Python
• PyTorch
• matplotlib
• Git
• GAN
• GenAI

Machine Learning Models and Tasks

Implementated from scratch supervised learning (Linear Regression, Logistic Regression, Neural Network, SVM) and unsupervised learning (K‑means, K‑medoids, GMM, PCA) models to perform different tasks.

• Python
• NumPy
• pandas
• matplotlib
• scikit-learn
• Supervised Learning
• Unsupervised Learning

Artificial Intelligence Algorithms and Problems

Designed and implementated from scratch state space search problems, local search problems, adversarial search problems and constraint satisfaction problems using Artificial Intelligence algorithms with object‑oriented programming and logic programming.

• Python
• NumPy
• Prolog
• Agent-Based Model
• Logic Programming