People sometimes raise an eyebrow when they see my CV. A degree in Theoretical Physics. A master's in AI. Another master's in Astrophysics. Three years as a Data Scientist and AI Engineer in industry. And now a PhD in deep learning. It's not the straightest line.

But when I look back at it, every step was necessary. Here's what I learned along the way — and what I'd tell anyone considering a similarly winding path into AI research.

Starting with Physics

I studied Theoretical Physics at Universitat de Barcelona from 2014 to 2019. People assume physics is a detour on the way to AI. I'd argue it's one of the best foundations you can have.

Physics teaches you to think in terms of systems, constraints, and approximations. You learn to build minimal models that capture the essence of a phenomenon without overfitting to its details. You get comfortable with mathematics not as a tool, but as a language for describing the world. These habits of mind transfer directly to machine learning research.

Two Masters, Two Perspectives

After my physics degree, I pursued a Master's in Intelligent Interactive Systems at UPF. This was my formal entry into AI — neural networks, NLP, computer vision, the works. It gave me technical depth in the field I wanted to work in.

I then did a second master's in Astrophysics and Cosmology at UAB. Some people thought this was a step backward. It wasn't. Working on cosmological data — huge, noisy, high-dimensional — forced me to think carefully about statistical rigor, model assumptions, and the difference between fitting data and understanding it. These are lessons that have made me a better ML researcher.

Three Years in Industry

Before starting my PhD, I spent three years working as an AI Engineer and Data Scientist — first at DRIVING01, building NLP conversational agents for companies like Liceu and real estate firms, then at Schneider Electric as a global AI consultant.

Industry taught me what research never quite can: what it feels like when your model fails at 2am on a production system with real users depending on it.

That experience changed how I think about robustness, reliability, and the gap between benchmark performance and real-world utility. It's directly relevant to my current research on model evaluation.

The PhD Decision

After industry, I missed the long-form thinking that research allows. I wanted to go deep on a problem instead of shipping quarterly. The opportunity to join UPF's IMVA group with Gloria Haro — working on something at the intersection of computer vision, audio, and self-supervised learning — felt like the right combination of intellectual depth and practical relevance.

The PhD has been everything I hoped. A visiting stint at NYU's MARL lab in 2024, collaborating with Magdalena Fuentes, broadened my network and led to an ICASSP publication. The research questions are genuinely hard and the answers genuinely matter.

What the Unconventional Path Gave Me

If you're considering a winding path into AI research, here's what I'd say:

  • Depth in any rigorous field transfers. Physics, mathematics, statistics, engineering — they all teach you to think carefully. That skill is rare and valuable in ML.
  • Industry experience makes you a better researcher. You develop instincts for what matters and what doesn't. You know which benchmarks to distrust.
  • The "conventional" path is not the only path. Many of the most interesting researchers I've met came from unexpected places. The field rewards curiosity and rigor more than credentials.
  • It's never too late to start. I was 27 when I started my PhD. That's not late — that's experienced.

If you're on a similar journey and want to talk about it, reach out. I'm always happy to chat.