cv

DPhil in Autonomous Intelligent Machines and Systems

University of Oxford

• DPhil on the fundamental properties of generative models and communication protocols for LLMs.
• Funded by a 250k USD studentship from Microsoft Research.

MSc Artificial Intelligence

University of Bologna

• GPA: 29.73/30, top of the class (170 students) Graduation grade: 110 cum laude.

Excellency Degree

Superior College of Bologna (Institute of Superior Studies)

• 7 STEM students are admitted each year and are provided with a personal tutor, exemption from tuition fees, free accommodation and an annual scholarship throughout their Bachelor's and Master's degrees.
• Yearly requirements: 28/30 GPA, no grade lower than 24/30, must take 6 additional excellency courses.
• GPA: 29.75/30

BSc Computer Science

University of Bologna

• GPA: 29.77/30, top of the class (150 students). Graduation grade: 110 cum laude.

• Samuele Marro, Emanuele La Malfa, Jesse Wright, Guohao Li, Nigel Shadbolt, Michael Wooldridge, Philip Torr, A Scalable Communication Protocol for Networks of Large Language Models.

• Tobin South, Samuele Marro, Thomas Hardjono, Robert Mahari, Cedric Deslandes Whitney, Dazza Greenwood, Alan Chan, Alex Pentland, Authenticated Delegation and Authorized AI Agents. ICML 2025.

• Samuele Marro, Davide Evangelista, Angelo Huang, Emanuele La Malfa, Michele Lombardi, Michael Wooldridge. Language Models Are Implicitly Continuous. ICLR 2025.

• Samuele Marro, Michele Lombardi. Computational Asymmetries in Robust Classification. ICML 2023

Head

Institute for Decentralized AI

• Research institute focused on security and infrastructure for decentralized and distributed AI systems.
• Two main units: Oxford and Stanford.
• Funded by the AI Safety Fund and the Cosmos Institute

Predoctoral Research Fellow

University of Bologna

• Department fellowship on adversarial robustness under prof. Michele Lombardi.

Visitor

Oxford University

• Research visit to the Department of Computer Science to conduct research on the Theory of Mind of LLMs under Dr. Emanuele La Malfa and Prof. Michael Wooldridge.

Student Researcher

University of Bologna

• Developed, under prof. Michele Lombardi and Michela Milano, an adversarial defense (implemented in Python + PyTorch) which is robust against white-box attacks.
• Wrote a Julia interface to formally verify the effectiveness of the defense using MIPVerify.jl + Gurobi.
• Proved that training a robust ReLU model is Sigma2P-hard, with implications for adversarial defense design.
• Obtained a 170k core-hour grant for the CINECA supercomputer cluster.

Teaching Assistant

University of Bologna

• Taught the applied ML module of "Introduction to Machine Learning" for the Fall '22 semester.
• Performed research on the latent space of diffusion models under prof. Andrea Asperti.

Research Intern

MindIt Srl + Kantar

• Developed a Tensorflow LSTM implementation compatible with IBM CPLEX optimization algorithms.
• Introduced tools to mitigate numerical issues with non-linear components of LSTMs.
• Rewrote Facebook's FBProphet time series model in Tensorflow.

AI Safety Fund Grant, 450k USD

• Grant by the AI Safety Fund (joint Google-OpenAI-Anthropic-Microsoft fund) on the decentralized oversight of agent networks.

Cosmos Institution Building Fellowship

• Fellowship by the Cosmos Institute to build new institutions for decentralized AI.

Microsoft Studentship, 250k USD

• Studentship for research into communication protocols between LLMs and multimodal memory in generative models.

CINECA HPC Grant

• Grant for the Galileo100 HPC cluster for research on the computational complexity of adversarial attacks.

Decentralized Oversight of Agent Networks

• Presented at the ERA Technical AI Governance Forum.

Agora: A Scalable Protocol for Decentralized Agent Communication

• Presented at Microsoft Ltd (2024) and Camel AI (2025).

The Case for Decentralized Intelligence

• Presented at the Oxford Maths Institute.

Computational Asymmetries in Robust Classification

• Leonardo Lesmo Award talk at AIxIA 2023.

Practical Machine Learning

• Presented at the University of Bologna Algo-UniBo organization.

Solidity Developer and Auditor

Freelance

• Performed Smart Contract audits, Web3 integrations and secure Smart Contract development.
• Specialized in EVM and EVM-like (e.g. Vite) security standards.

Tutor

Liceo Scientifico L. Cocito, Alba (Italy)

• Provided one-on-one and one-on-two tutorship for students struggling with Mathematics and Physics.

Independent Research

• Extended Grammatical Evolution's Backus-Naur Forms to support runtime checks and automatic correction of invalid production rules.
• Implemented this variant of Grammatical Evolution in C# and Python.

Intern

Sistemi Tre, Alba (Italy)

• Automated database queries through the in-house software interface using AHK, leading to a 3-4x increase in data retrieval speed.
• Assisted IT staff during on-site technical support.

Gitcoin Round 13 Hackathon, 6k USD bounty

• Wrote four smart contracts in Soliditypp for the Vite blockchain.
• Wrote unit tests for such contracts using Vuilder and Javascript.

GreenNFT Hackathon, 3k USD bounty

• Created a model to estimate the environmental impact of individual transactions on Ethereum.
• Identified techniques to reduce a user's individual impact by 70% with the help of Python scripts.
• Developed Green NFT Helper, a JavaScript + Svelte.js website to help users to reduce their emissions.

Gitcoin Round 9 Hackathon, 1k USD bounty

• Developed a Chrome extension in JavaScript to search on IPFS directly from the omnibox.

Ancalau Award, 3k EUR award

• Trained in Python + Tensorflow a CNN to identify illegal dumping locations from aerial data.
• Deployed the model as part of an Android app (written in Java) to capture aerial data and relay landfill locations in real time to a quadcopter.