Applied AI Scientist at Mistral
I’ve joined Mistral AI as an Applied AI Scientist & Research Engineer.
Welcome on Florian Wiesner
Recent content in Welcome on Florian Wiesner
Latest posts
Formalized Proofs in Physics
Exploring formal methods in physics.
Towards a Physics Foundation Model
With this work, I wanted to explore whether the concept of foundation models (LLMs for language) can be applied to physics simulations. The idea is to train a single model on a large corpus of diverse simulation data, such that it can generalize to new physical systems without retraining. I also wrote a blog post explaining the work in more detail.
Enhancing Bubble Removal in Geometry-Optimized Electrodes
This work demonstrates that geometry-optimized 3D-printed lattice electrodes achieve better electrochemical performance than conventional stochastic foams in alkaline water electrolysis, despite having lower surface area. The key innovation is the Venturi-like entrainment mechanism that improves bubble evacuation through straight flow channels.
Physics Foundation Model Talk on Youtube
I gave a talk on physics foundation models for the StartUp First Principles, now available on YouTube .
Dissertation done!
I successfully defended my dissertation with highest honors, earning the title of “Dr.-Ing.” or PhD.
Using Agents to Assist Research Tasks
I am starting to use AI agents to do the research for me, and here is my experience so far.
Data Mining for Enhanced PEM Electrolysis
The main goal of this work was to tackle the reproducibility crisis in PEM electrolysis research. Many studies focus only on the catalyst or other subparts of the cell and do not report all the parameters that influence the performance. We created a database with more than 1,000 samples from 127 publications to get more insights into the influence of different parameters on the performance.
Recap: More than 20 students supervised
I am revisiting my experience supervising more than 20 students in their research projects.
Towards a Physics Foundation Model
A new transformer-based model that learns multiple physical systems and generalizes to unseen scenarios.
ML Training Boilerplate Project
This project serves as a boilerplate for training machine learning models.
Merck Innovation Cup 2025
Participated in the Merck Innovation Cup 2025, developing an innovative solution for multi-modal models in material discovery.
Physics Foundation Model Research at University of Virginia
Excited to announce the start of my research stay working on Physics Foundation Models at the University of Virginia!
Unveiling the Role of PTFE Surface Coverage on Controlling Gas Diffusion Layer Water Content
Accurate prediction of water content in fuel cells or electrolyzers is crucial for their performance. We developed a new algorithm which is fast and accurate (enough). I am not sure if this simulation will be used in industry, but it opened up new questions and challenges regarding the role of PTFE and where to place it.
Spatio-Temporal Electrowetting and Reaction Monitoring in Microfluidic Gas Diffusion Electrode Elucidates Mass Transport Limitations
We developed a novel experimental setup (microfluidic) to visualize and study the complex interactions between water and electrodes in CO₂ reduction reactors. We hope to find new insights into the behavior of such gas diffusion electrodes.
Hiking
If I find the time, I love to go hiking. Recently, my wife and I visited the US west coast and enjoyed Yosemite National Park (photo) and Zion National Park. We also got to see the gigantic sequoias in the Sierra Nevada mountains as well as the Grand Canyon. Last year, we went to the Alps, specifically the Königssee, which offers breathtaking views of the lake and the surrounding mountains. During…
Snowboarding
Most winters, I try to get at least one week of snowboarding in. I’d say I’m an allrounder, enjoying some carving as well as freeriding in powder or some smaller jumps (not very good yet!). With the right people, snowboarding is one of my favorite activities. Here, you can see me enjoying the slopes of Kleinwalsertal while a good friend of mine is recording. As a teenager, I really…
Software and Coding
I am not only a software developer at work, but also enjoy working on personal projects. You can find some of my more polished projects on my github, but oftentimes I also just play around with new technologies. Here you can see a small animation I created with a custom n-body solver. It shows two suns orbiting around each other, stealing each other’s planets. Currently, I put most of my…
Custom CUDA Kernel
A custom CUDA kernel for masked 3D convolution operations.
N-body Simulation
A simple N-body simulation implemented in C++.
Additive Manufacturing of Intertwined Electrode Pairs - Guided Mass Transport with Gyroids
During this study, I had access to a metal 3D printer. I used it to create novel electrode designs based on mathematical surfaces called triply periodic minimal surfaces (TPMS). Such 3D electrodes can have significantly higher surface area than conventional parallel plate electrodes as well as improved mass transport. Additionally, it is just fun to create crazy-looking metal things!
Joined the TriggerInk project
Became a member of the TriggerInk project focusing cartilage tissue engineering using 3D bioprinting.
Beginning of PhD
Started my PhD journey in Chemical Engineering at RWTH Aachen University.
Process model for high salinity flow-electrode capacitive deionization processes with ion-exchange membranes
This was my Bachelor thesis. I developed a model (in gProms) for the flow-electrode capacitive deionization (FCDI) process. FCDI is a novel water desalination technology our chair uses to recover lithium from brines (battery waste). The simulation is able to predict the process behavior for a wide range of salinities.