About Me:

My name is Raphael Pisoni. I am a Machine Learning Researcher but my path was not a straight line.

The Detour

My journey started in Brixen, a small city in northern Italy. I’ve always been driven by STEM, which initially led me to Vienna to study Environmental Engineering and Water Management at the University of Natural Resources and Life Sciences (BOKU).

Close to finishing my bachelor's, I was hit by a serious illness that forced me to return to my hometown for recovery. It was a not an easy time, but it became the defining moment of my career. Confined to a hospital bed, I started experimenting with Machine Learning and Neural Networks.

It was 2013. The deep learning boom was just beginning, and although I knew no one seriously involved in the field, I was hooked. I decided to enroll in Computer Science courses at the Free University of Bolzano nearby to build the theoretical computer science foundation I needed to turn my curiosity into code.

The Foundation

At the University, I found a mentor in Professor Tammam Tillo. He supported my interest in Deep Learning and Computer Vision and offered me a desk in a PhD office, even though I was just an undergraduate.

With his help, I dove into depth estimation using neural networks. I wrote two papers and a bachelor thesis, aiming high with submissions to NeurIPS and CVPR. Back then, none of them were accepted despite reaching strong results, which was a humbling lesson in the rigors of academic research.

However, that persistence paid off. Fast forward to today, and I am proud to have published work in several top-tier venues.

The Engineer

Between those early academic attempts and my current research, I spent years in the trenches of industry.

At Partium.io: I worked as a Senior ML Engineer, rewriting training pipelines from scratch to cut training time from 7 days to under 14 hours at increased performance, and spearheading innovative multimodal search solutions that work well across modalities and languages.

At Tensoreye (acquired by Zenseact/Volvo): I led the transition to mixed-precision training for autonomous driving models, speeding up training by up to 60% and optimizing performance across thousands of GPUs.

This blend of theoretical curiosity and engineering pragmatism defines my work today. I don't just want to know why a model works; I want to make it work efficiently, at scale.

Current Focus

Today, I work as a Researcher and Senior Data Scientist at SCCH, where I focus on:

• Representation Learning
• Multimodal & Multilingual Learning
• Unsupervised- and Self-Supervised Learning
• The Geometry of Neural Networks (specifically Space Folding)

This blog is where I document my journey. You’ll find posts about my latest research, some deep dives and the "fun" side projects I build along the way to keep learning.

If you'd like to chat about research, engineering, or anything in between, feel free to drop me an email or connect on LinkedIn.