←
Fabian Zills70 posts
Fabian Zills avatar

Fabian Zills

@PythonFZ
◷ Joined May 2014▧ 7 Photos and videos
Potential Sciences avatar
⟳ Fabian Zills retweeted
Potential Sciences@potentialsci_·Nov 17

The bottleneck in atomistic foundation models has moved from model fit to data quality: models trained on OMol25 (2025) now sit closer to their DFT labels than those labels sit to the gold standard. Our first technical lab note: potentialsciences.com/blog/2…

Post attachment
Potential Sciences avatar
⟳ Fabian Zills retweeted
Potential Sciences@potentialsci_·Nov 6

A deep, technical chat between Hatem and @CorinWagen: simulated data at coupled-cluster accuracy, lead optimisation, the hardware lottery and building in the UK with @uksovereignai. Thanks @RowanSci for having us!

Yesterday, Potential Sciences came out of stealth to build a UK frontier lab for chemistry. I spoke with Hatem Helal, Potential's CEO, about what they're building and why their simulation-first approach complements other AI-for-science teams. rowansci.com/blog/a-conversa…

Fabian Zills avatar
Fabian Zills@PythonFZ·Nov 5

Excited to finally share that I joined Potential Sciences 6 months ago to build new methods and software for the next generation of AI in computational chemistry.

AI for drug discovery has learned almost everything the existing experimental record can teach it. The next leap needs new data, and the laws of physics can supply it. Potential Sciences is out of stealth to solve this problem, backed by @uksovereignai and @Twinpathvc.

0:17
Fabian Zills avatar
Fabian Zills@PythonFZ·26 Nov 2025

I was looking into how AI detection in text works. It's quite simple, you can e.g. use GPT2 to get the probability on the next word in the given context. I put that into a textual app github.com/PythonFZ/findllm (The code is 99% generated by claude agents - no warranties!)

0:12
🦉DVC avatar
⟳ Fabian Zills retweeted
🦉DVC@DVCorg·28 Oct 2025

@MoritzRSchfer1, @PythonFZ, and team introduce APAX, a high-performance JAX-based framework for building machine-learned interatomic potentials with active learning. Learn more about APAX and how they use DVC in their workflow. 👇🏽 hubs.la/Q03Pkljm0

Post attachment
Fabian Zills avatar
Fabian Zills@PythonFZ·19 Sep 2025

Which foundation MLIP is best suited for your application? The MLIPX framework can help you answering this question. Checkout the Paper iopscience.iop.org/article/1… and github.com/basf/mlipx #MLIPX

Moritz R Schäfer avatar
⟳ Fabian Zills retweeted
Moritz R Schäfer@MoritzRSchfer1·31 Jul 2025

New Publication 👏 We are excited to share our latest paper, "Apax: A Flexible and Performant Framework for the Development of Machine-Learned Interatomic Potentials", published in the JCIM. Read the full paper: pubs.acs.org/doi/10.1021/acs… Code: github.com/apax-hub/apax

Edvin Fako avatar
⟳ Fabian Zills retweeted
Edvin Fako@efssh·17 May 2025

Started with some slides for AutoAdsorbate-ended up with an animated deep dive into surface site detection. It works on any surface, no bulk symmetry needed. Here’s what this one-liner really does: s=Surface(slab211) code: github.com/basf/autoadsorbat… paper: tinyurl.com/yfbxnbds

0:26