Hi there!
I work at Pasteur Labs on Tesseract, an open-source tool for wrapping simulation components so they compose into end-to-end differentiable pipelines (w/ some existing Julia examples via juliacall). We’re running a month-long online hackathon throughout August around that idea, and we figured a few people here might find it interesting.
The challenge is to take differentiable components that don’t naturally fit together (using different languages, frameworks, or autodiff strategies) and compose them so gradients flow through the whole thing, then use that to solve a real design, inference, or training problem. Tracks cover inverse design, multi-physics coupling, hybrid ML + mechanistic models, differentiable inference/UQ, and differentiable rendering.
It’s free, solo or teams up to 4, runs Aug 3–31. There’s a $20k prize pool and we’ll help the strongest entries turn their work into a paper, while we hope it will provide the community with invaluable insight on the composition problem itself.
I’m not allowed to post links here but if this sounds interesting you’ll find all the details if you google "tesseract hackathon 2026". Happy to answer any questions here, too.