Hi, I'm Pietro!

I'm a PhD candidate at the University of Queensland (opens in a new tab), working with Prof. Gianluca Demartini (opens in a new tab), Dr. Joel Mackenzie (opens in a new tab), and Dr. Kevin Roitero (opens in a new tab).

My research broadly focuses on understanding bias and information representation in large language models (LLMs), with particular interests in political and perspective bias, relevance assessment, multilinguality, and model internals. Outside of my specific research interests, I’m mostly just very curious and enjoy learning about things I know little about.

Publications

Selected research

01
IP&M 2027
Political Ideology Shifts in Large Language Models
Pietro Bernardelle, Stefano Civelli, Leon Fröhling, Riccardo Lunardi, Kevin Roitero, and Gianluca Demartini
Information Processing & Management, vol. 64, no. 2, article 105124, February 2027
DOI (opens in a new tab)arXiv (opens in a new tab)
Across seven open-weight models and 200,000 synthetic personas, shows that persona conditioning and explicit ideological cues shift models' political-compass positions, with larger models covering a broader ideological range.
```
@article{bernardelle2027political,
  author = {Bernardelle, Pietro and Civelli, Stefano and Fr\"{o}hling, Leon and Lunardi, Riccardo and Roitero, Kevin and Demartini, Gianluca},
  title = {Political Ideology Shifts in Large Language Models},
  journal = {Information Processing \& Management},
  publisher = {Elsevier},
  volume = {64},
  number = {2},
  pages = {105124},
  month = feb,
  year = {2027},
  issn = {0306-4573},
  doi = {10.1016/j.ipm.2026.105124},
  arXiv = {2508.16013}
}
```
02
CIKM 2026
LLMs Encode Relevance as a Layer-Wise Cross-Lingual Signal
Pietro Bernardelle, Samaneh Mohtadi, Stefano Civelli, Joel Mackenzie, and Gianluca Demartini
Proceedings of the 35th International ACM Conference on Knowledge and Information Management (CIKM)
arXiv (opens in a new tab)
Shows that query–document relevance is linearly decodable from LLM activations as a depth-dependent signal, and tests how well it carries across languages.
```
@inproceedings{bernardelle2026internalsrelevance,
  author = {Bernardelle, Pietro and Mohtadi, Samaneh and Civelli, Stefano and Mackenzie, Joel and Demartini, Gianluca},
  title = {LLMs Encode Relevance as a Layer-Wise Cross-Lingual Signal},
  booktitle = {Proceedings of the 35th International ACM Conference on Knowledge and Information Management (CIKM)},
  arXiv={2607.15555},
  year = {2026}
}
```
03
CIKM 2026
Persona Conditioning as an Assessor-Sensitivity Probe for LLM-Based IR Evaluation
Samaneh Mohtadi, Pietro Bernardelle, Joel Mackenzie, and Gianluca Demartini
Proceedings of the 35th International ACM Conference on Knowledge and Information Management (CIKM)
arXiv (opens in a new tab)
Uses persona conditioning to probe LLM relevance assessors, finding that judgments shift in strictness and emphasis rather than flipping, and that smaller models amplify the resulting instability in system rankings.
```
@inproceedings{mohtadi2026personarelevance,
  author = {Mohtadi, Samaneh and Bernardelle, Pietro and Mackenzie, Joel and Demartini, Gianluca},
  title = {Persona Conditioning as an Assessor-Sensitivity Probe for LLM-Based IR Evaluation},
  booktitle = {Proceedings of the 35th International ACM Conference on Knowledge and Information Management (CIKM)},
  arXiv={2608.10385},
  year = {2026}
}
```

View all research

Experience and education

Resume

Jul 2024 – Present
PhD, Artificial Intelligence
The University of Queensland
Alongside
- Feb 2025 – Nov 2025
  Academic Tutor
  UQ School of EECS
  Data Mining (INFS4203), Information Retrieval and Web Search (INFS7410), Data Analysis at Scale (DATA7201)
Feb 2023 – Jul 2024
M.Sc. in Engineering Science (Management)
The University of Queensland
Distinction Dean's Commendation for Academic Excellence · Semester 1, 2024
Alongside
- Nov 2023 – Jan 2024
  Research Assistant
  UQ School of EECS
  Graph-based reinforcement learning for pair selection
- Aug 2023 – Nov 2023
  Research Intern
  CIRES
  Graph-based reinforcement learning for pair selection
Sep 2021 – Dec 2024
M.Sc. in Computer Science and Engineering
Politecnico di Milano
Sep 2018 – Jul 2021
B.Sc. in Engineering of Computing Systems
Politecnico di Milano

Download CV

Political Ideology Shifts in Large Language Models

LLMs Encode Relevance as a Layer-Wise Cross-Lingual Signal

Persona Conditioning as an Assessor-Sensitivity Probe for LLM-Based IR Evaluation

PhD, Artificial Intelligence

Academic Tutor

M.Sc. in Engineering Science (Management)

Research Assistant

Research Intern

M.Sc. in Computer Science and Engineering

B.Sc. in Engineering of Computing Systems