teaching

Courses, tutoring, supervised projects, and thesis co-supervision at the University of Bologna.

courses

Bologna Business School

Text Mining and Natural Language Processing

Adjunct Professor, co-instructor with Gianluca Moro A.Y. 2023–2026

Master in Data Science and Business Analytics, Bologna Business School. I teach the laboratory module (12 hours) and design and grade the coding assignments. Student satisfaction 4.5/5 (2024–25) and 4.3/5 (2023–24).

University of Bologna

Big Data Analytics and Text Mining, Text Mining and LLMs module

Tutor, course led by Gianluca Moro A.Y. 2023–2026

M.S. in Artificial Intelligence, University of Bologna. I run the labs (LLM prompting and fine-tuning, graph neural networks and knowledge graph embeddings, relational deep learning) and co-supervise the exam projects listed below.

thesis co-supervision

62 Bachelor's and Master's theses on NLP and deep learning co-supervised at the University of Bologna, counted daily from AMS Laurea, the university's open-access thesis repository.
# theses per year appear after the first daily update
Hover a year for the Bachelor's and Master's split.

Selected theses

  • On-Device Retrieval-Augmented Generation over Personal Clinical Records Ferri, A. (2026). Paper accepted at KEIR @ CIKM 2026
  • Automatic Optimization of Agentic Pipelines for Protein Function Prediction Rubri, M. (2026)
  • Graph-of-Mark: Graph-Based Visual Prompting for Enhanced Spatial Reasoning in Multimodal Language Models Buzzoni, M. (2026). M.S. thesis, Artificial Intelligence. Paper at AAAI-26 (oral)
  • Multimodal Generative Information Retrieval of Chest X-Rays Grounded on ICD-9 Taxonomy Raponi, M. (2025). B.S. thesis, Computer Science and Engineering
  • End-to-End Extraction and Injection of Graphs: A Self-Supervised Neuro-Symbolic Method for Explainable Large Language Models Zecca, A. (2024). M.S. thesis, Artificial Intelligence
  • Large Action Models: End-to-End Retrieval-Enhanced Learning for Generating Function Calls from Instruction Manuals Monaldini, N. (2024). B.S. thesis, Computer Science and Engineering. Paper at EMNLP-25 (PORTS)
  • Leveraging Large Language Model Distillation to Enhance Zero-Shot Named Entity Recognition and Classification Cocchieri, A. (2023). M.S. thesis, Artificial Intelligence. Papers at NAACL-25 and ACL-25
  • To Generate or to Retrieve: On the Effectiveness of Artificial Contexts for Biomedical Question Answering Presepi, A. (2023). B.S. thesis, Computer Science and Engineering. Paper at ACL-24 (MedGENIE)

supervised exam projects

A selection of the exam projects (6 CFU) and project works (3 CFU) I co-supervised in Big Data Analytics and Text Mining, a course attended by more than 80 students per year. Each topic is agreed with the student, and several grow into theses and papers.

A.Y. 2025–2026 (a selection)

Protein function prediction with geometric GNNsGearNet, ESM-GearNet and GGN-GO on CAFA-5 using 3D structures from AlphaFold-DB.Project + Project Work
Uncertainty quantification for LLMs, a clinical tutorialWhite-box, black-box and multimodal uncertainty with lm-polygraph and uqlm, with calibration and benchmarking.Project + Project Work
DSPy tutorial on a multi-agent ICD coderRe-implementing the Code Like Humans pipeline with DSPy and open models served by llama.cpp.Project
Biomedical NER with reinforcement learningMedGemma fine-tuned with LoRA SFT and GRPO to extract typed entity spans as JSON.Project + Project Work
First-order-logic reasoning in small LLMsPiCSAR answer selection and logical-equivalence evaluation for FOL proof chains checked with Prover9.Project
Compositional faithfulness of text-to-image modelsA benchmark of five diffusion models on T2I-CompBench and GenEval, scored with OWLv2.Project + Project Work
Synthetic clinical documentsA LangGraph multi-agent pipeline producing realistic, degraded clinical PDFs with LLM-as-a-judge checks.Project Work
Chest X-rays generated from radiology reportsClinical and anatomical faithfulness of report-conditioned diffusion models on MIMIC-CXR, with controlled editing.Project + Project Work
Multimodal ICD coding with vision-language modelsFine-tuning a multimodal LLM with a custom tokenizer, and agentic use of the same VLM on ChestMNIST.Project + Project Work
Robot learning with LeRobotDiffusion policy and ACT in simulation, and a comparison of SmolVLA and X-VLA on LIBERO.Project
A personal chess coachOpening repertoires tailored to a player's games, repertoire gaps, and move commentary from Stockfish and an LLM.Project
Clinical relational deep learningExtending RelBench with MIMIC and eICU databases and new temporal prediction tasks.Project
Person search with vision-language modelsFinding a person in image collections from a natural-language description.Project
Few-shot summarizationComparing small fine-tuned models and larger prompted LLMs on news summarization with little data.Project
Backdoor attacks through quantizationWhether a model that behaves well at full precision can hide malicious behaviour that appears once quantized.Project
Data contamination in LLM benchmarksDetecting when test questions have leaked into a model's training data.Project
Visual question answeringPrompting and fine-tuning strategies for multimodal LLMs answering questions about images.Project
Faithful explanations for text classifiersCounterfactual tests of whether a classifier relies on style or on topic.Project Work

international thesis committees

  • Universidad Técnica Federico Santa María (UTFSM), Valparaíso, Chile (2026). External member of the thesis committee for the Magíster en Ciencias de la Ingeniería Informática: written evaluation of the dissertation and participation in the oral defense (counterfactual explanations, explainable AI, NLP).