Marcello Politi

Research Scientist at the Ethereum Foundation, based in Rome, Italy

Working on multi-agent systems, AI safety, and the new agent economy — building the primitives for decentralised AI and the agentic web on Ethereum.

Email / LinkedIn / Website / GitHub

👩🏼‍💻 Technical Experience

Research Scientist @ Ethereum Foundation (Oct 2025 - Present)

Working on multi-agent systems, AI safety, and the new agent economy.
Contributing to core primitives for decentralised AI on Ethereum, including the ERC-8004 agent standard and verifiable inference.
Researching LLM jailbreaking in strategic economics games, with applications to AI security in the agentic economy.

Deep Learning Engineer @ Pi School (Jan 2023 - Oct 2025)

Project lead for ESA-SatcomLLM (€400K, 9 months), directing technical development and managing a team to build an open-source LLM for Satcom ARTES.
LLM Engineer for ESA-Eve, leading fine-tuning and RAG implementation in collaboration with Mistral AI and ESA Φ-lab; presenter at ESA-NASA AI FM Workshop and LPS 2025 session co-organiser.
Contributor to multiple successful grant proposals totalling over €46M.

Co-founder and Head of AI @ BrianknowsAI (Sep 2023 - Feb 2025)

Co-founded and developed Brian, an AI-powered Web3 agent enabling non-custodial transactions, smart contract deployment, and blockchain data access via prompting.
Won multiple hackathon awards (EthPrague, AugmentHack) and was invited to present at the decentralised EO event at ESRIN, co-organised by ESA and Swarm.
Secured $150K pre-seed funding and raised $100K+ in grants, with 10k DAUs.

Young Graduate Trainee in Software Engineering @ ESA (Sep 2021 - Sep 2022)

Contributed to the optimisation of data access flows within ESA’s EO Common Services, focusing on improving end-user experience.
Worked on meta-systems to enhance software architecture, with additional research supporting ESA’s green computing objectives.
Participated in the Living Planet Symposium 2022 as part of ESA’s EO outreach and technical engagement efforts.

Deep Learning Research Intern @ INRIA (March 2021 - July 2021)
Conducted research on neural network compression through iterative pruning techniques, complementing thesis work on efficient deep learning. Implemented and evaluated pruning strategies using the Julia programming language, focusing on reducing model complexity while preserving performance for resource-constrained environments. Supervised by Emanuele Natale and Andrea Clementi.

Technologies used: Julia, Flux.

Web Analytics and Information Outreach Intern @ ESA (Sep 2020 - March 2021)

Gained hands-on experience in web data analysis by integrating analytics tools (e.g. Matomo) with custom scripts to support data-driven decision-making.
Applied semantic web technologies such as DBpedia and Schema.org for effective data and content linking.
Contributed to improving digital insights and strategy within ESA’s web platforms.

🗞 Other Experiences

Contributing Writer @ Towards Data Science / Medium / Red Hot Cyber
Online publications providing insights into machine learning, deep learning and AI security.

Selected articles:

Hackathon Organiser - “Mission Control TTX” (1st edition) @ Space Generation Advisory Council (SGAC) (April 2023 - Present)
The Space Generation Advisory Council is a global non-governmental, non-profit (US 501(c)3) organisation and network which aims to represent university students and young space professionals to the United Nations, space agencies, industry, and academia.

Dock3 - The Startup Lab @ Dock3 (Mar 2021 - Jun 2021)
Dock3Sprint is the incubation program of RomaTre University, where the teams go from idea to market through 15 workshops covering fundraising, growth hacking, lean metrics, lean management, legals, hardware production and more.

Hackathon 2019 - Organizer @ Tor Vergata (Feb 2019 - May 2019)

Website co-creator
Logistic
Development of competition themes

🏆 Accomplishments

Machine Learning DevOps Engineer @ Udacity Nanodegree (April 2024)

Best System Award @ EVALITA 2020
Development of a stance detector system for Italian tweets exploiting BERT with a transfer learning approach. Paper available here.

Hackathon winner @ ConfCooperative Hackathon (Nov 2019)
The github repo of the project. Health Cooperation application that allows user identification by eliminating paperwork, a review system of affiliated facilities and user profiling using data in order to seek solutions to provide tailored services.

🛠 Skills

Programming: Git, GitHub; Python, JavaScript, OOP; My/NoSQL, VectorDB; PyTorch, TensorFlow, Axolotl, Unsloth, HF, Langchain; MLflow, W&B; AI Agents, MCP, a2a

Project Management: Asana, Notion, Gantt Charts; Agile, Scrum; Microsoft 365, Google Workspace; Risk management; Financial Management

Soft Skills: Team worker; Adaptability; Fast learner; Social intelligence; Story telling; Presentation skills

💬 Languages

Italian: Native
English: Proficient user (C1)
Hungarian: Proficient user (B2)

👩🏼‍🎓 Education

Fellow at the School of Artificial Intelligence
Pi School - Rome, Italy (May 2022)

Advanced Deep Learning
Deep Learning Italia Academy / Experis Academy - Rome, Italy (April 2022)

Master in Computer Science - University of Tor Vergata, Rome, Italy (July 2021)
Summa cum laude
Thesis: “An Assessment of Iterative Pruning Methods for Artificial Neural Networks in Julia”

Advisor: Prof. Andrea Clementi
Co-Advisor: Prof. Emanuele Natale

Bachelor in Computer Science - University of Tor Vergata, Rome, Italy
Thesis: “Dynamic of Bitcoin Network: empirical analysis of a full-node’s neighborhood”

Advisor: Prof. Francesco Pasquale
Available here