Hello, I'm Flavio 👋🏻

I'm a Machine Learning Engineer based in Switzerland. I'm passionate about all things AI. I train and deploy cutting-edge models with a focus on self-supervised learning, and NLP applications. Let’s create something impactful!

About Me

flavio:$
I earned a Master’s degree in Electrical Engineering from ETH Zurich, one of the world’s leading institutions for science and technology, where I specialized in Machine Learning and Signal Processing. My research focuses on expanding the capabilities of transformer architectures across diverse applications, including sequence modeling and computer vision. I have hands-on experience in representation learning, particularly with Implicit Neural Representations (INRs) and leveraging transformers to extract meaningful representations across various domains. My work has been published at NeurIPS.
Beyond AI, I have a strong passion for blockchain technology and cryptography, actively exploring applications at the intersection of distributed computing and machine learning.

My Skills

flavio:$
flavio:./skills$

Frameworks

PyTorch
TensorFlow
HuggingFace
Pandas
Numpy
Scipy
Scikit-Learn
Flask

Coding

Python
C++
Java
JavaScript
HTML
CSS
SQL
Rust
Solidity

Dev-OPs

Docker
Kubernetes
AWS

Languages

English (Fluent)
German (Native)
French (Intermediate)

My Projects

flavio:$
flavio:$

Abstractions behind Integer Sequences

My research explores how AI models understand integer sequences and their underlying logic. As part of my bachelor’s thesis, I collaborated with two Ph.D. students from ETH Zurich to develop an enhanced Transformer model that improves AI’s capacity for mathematical and logical reasoning, an area where existing models struggle. Our adapted architecture was designed to learn the abstractions within integer sequences, bridging the gap between language understanding and mathematical reasoning. This work has been published at NeurIPS, one of the most esteemed conferences in AI. To support our research, we created a comprehensive benchmark and dataset, along with a full suite of tools for training and testing models on integer sequences.

Paper Codebase
flavio:$
flavio:$

Super Resolution with Neural Representations

As a Research Assistant at ETH Zurich, I advanced research on super-resolution tasks using Implicit Neural Representations (INRs), building upon my Master's thesis work. Unlike traditional digital methods that store data in discrete grids with fixed resolutions, INRs model data as continuous functions, enabling sampling at arbitrary resolutions. This approach is particularly well-suited for super-resolution tasks. To further explore their potential, I trained a Vision Transformer model on diverse data modalities, demonstrating the versatility of INRs. Our findings are on track to be published at ICCV, one of the leading AI conferences.

Codebase
flavio:$
flavio:$

LLM for knowledge retrieval

Together with two fellow ETH graduates, we developed a comprehensive knowledge retrieval system from scratch for the Swiss bank Julius Baer, leveraging large-language models fine-tuned on internal documents. This system enhances employee productivity by enabling fast and efficient access to company information and policies while ensuring compliance. Built with Python and PyTorch, it seamlessly integrates into daily workflows. To further optimize document retrieval, we constructed an embedding space, enabling the model to provide accurate source citations when delivering information.

Julius Bär
flavio:$
flavio:$

Web Design and Frontend Development

In addition to developing AI projects in Python, I have a strong interest in frontend development with HTML, CSS, and JavaScript. This passion has led to several web design projects, some dating back to my high school years. Notably, I built my own website entirely from scratch using only HTML, CSS, and JavaScript, without any additional frameworks or libraries.
My first complete web development project was in 2014 for Thurnheer Obst, a small family-owned agricultural company specializing in fresh local fruits. Although I no longer maintain the site, it marked my introduction to web design. I later contributed to the web design for airlinetools, a company that developed a sophisticated web-based Airbus A320 simulator built by professional pilots.
Most recently, I designed and currently maintain the website for Avanox AI, an AI startup where I am also a co-founder. At Avanox AI, we simplify AI-driven automation for businesses of all sizes.

Thurnheer Obst airlinetools Avanox AI
flavio:$
flavio:$

Blockchain and Cryptocurrencies

I have a deep passion for cryptography, blockchain technology, and cryptocurrencies. My interest in distributed computation was sparked during my master's studies through multiple courses from my supervising professor, whose work on improving the Bitcoin Lightning Network was both inspiring and influential. As part of this passion, I developed a cryptocurrency trading bot that directly interacts with the Solana and Ethereum blockchains, utilizing Node.js, Rust, and Solidity. Additionally, before Ethereum’s transition to Proof of Stake with ETH 2.0, I built and sold various Ethereum mining rigs.

Trading Bot