🙏 Thanks to my collaborators @jla-gardner.bsky.social Louise A. M. Rosset Andrew L. Goodwin @vlderinger.bsky.social 💃 🕺
And to our funders: @novo-nordisk.bsky.social @erc.europa.eu @ukri.org 🙏

@ox.ac.uk | @oxfordchemistry.bsky.social | @somervillecollege.bsky.social | DTU | DTU Energy

I’m excited about the potential: helping the scattering community move toward reliable automation, and supporting autonomous labs in making real-time decisions beyond pre-trained ML models.

👉 Full preprint: arxiv.org/abs/2510.05938

Applications span molecules 🧬, crystals 💎, nanoparticles ⚪ & amorphous matter 🌫️. Our method even reveals when multiple atomic structures give identical scattering — and shows when more experimental input is needed in autonomous labs (Figure 2).

We propose a new approach: a differentiable optimisation framework that unifies scattering 📊, energetics, & chemical constraints. Instead of relying on training data, it directly refines candidate structures against experiments.

⚠️ But there’s a catch: ML models are inherently bound to their training data, making them unreliable for uncharted chemistries — exactly where discovery happens.

⏳ In my research I have built ML methods to automate this process. ML can map structures to scattering patterns and deliver split-second interpretations — enabling self-driving experiments where synthesis, measurement, & analysis are connected in a closed loop 🔁.

For over a century, X-ray ✨ and neutron ⚛️ scattering have been central to chemistry & physics. Yet interpretation remains a bottleneck — still reliant on manual expert refinement.

Thanks to @ox.ac.uk, Oxford Chemistry, @somervillecollege.bsky.social, DTU - Technical University of Denmark, DTU Energy, Novo Nordisk Foundation for making it possible 🙏

Huge thanks to the organisers and judges for this incredible opportunity and to the other awardees for inspiring me every step of the way. You are superstars! ⭐

I would love to join too ☺️