Federico Mazzoni
Hi! I'm an AI Engineer with a PhD background and experience in deploying LLM-based systems and automation pipelines. I currently work at Domotz, integrating new AI solutions within the company's framework. On this page, you'll find the projects I've undertaken over the years, whereas here you can find my up-to-date CV.
Feel free to drop an email or add me on LinkedIn! I always love going out of my comfort zone and learning something new.
My Portfolio
Personal Projects
These are my more personal projects. :)
- StockAgent is a multi-agent AutoGen system that generates stock market insights. It includes a custom MCP server using yfinance.
- HINTT Is Not a Traditional Translator, but an interactive localisation tool for real-time text translation leveraging OpenAI visual capabilities.
- Remember those old black and white Game Boy games? Simple Palette Maker allows the user to colorize them when played on the Analogue Pocket.
Academic Projects
These are the projects I worked on during my PhD. The code might be a little messy and not up to my current standard.
- TETRA (TExtual TRust Analyzer) is a Python library to analyse conversational trust following three linguistic maxims. We made use of Sentence Transformers, Information Density detection, and existing libraries. Code.
- Frank is a Hybrid Decision-Making system, learning little by little from the user to help them make good, unbiased, and consistent decisions. It employs Incremental Learning in a novel way. Developed as part of the EU TANGO Project. Code.
- Bridget aims to “bridge the gap” between Hybrid Decision-Making systems where the users are in command (such as Frank) and those where the machine is in command (such as Learning-to-Defer). It dynamically switches between the two states.
- GenFair is a data augmentation framework. It employs a genetic algorithm to create synthetic data to make a dataset fair. Developed as my Master’s thesis. Code.
- FairNet is GenFair’s counterpart for networks. We assumed that in a “fair” network, no group should be marginalized, and we created a recommender system to achieve that condition. Code.