How do I join the hackathon?
- Join our Discord server (required).
- Enter your email on the home page and hit "Register for hackathon".
- Check your email for confirmation and next steps.
Can I participate solo?
Yes. Solo builds and teams of up to 4 are both welcome.
What's the prize?
A single $1,000 prize pool. One project wins the whole thing. All participants also get a certificate and a spot in the community gallery.
Do I need to pay to enter?
No. Registration and participation are free.
How do I submit my project?
Build your project in a
public GitHub repo created after Sep 10, 2026, make the first line of your README exactly:
This project was submitted to the ryze.ai hackathon by [NAME OF SUBMITTER].,
then use the
"Submit your project" form on the home page to submit your GitHub URL and contact email.
See
submission guidelines for full details.
Why does the repo need to be created after Sep 10?
This ensures all projects are built fresh during the hackathon. Repositories older than Sep 10 are automatically rejected.
Do you provide starter templates?
No. Bring your own stack. We provide resources and advice, not boilerplate.
Can I use a quantized model?
Yes, but the raw parameter count of the model architecture must be ≤500M. Quantization reduces file size, not parameter count. If you use a larger base model loaded in a pruned/quantized form that drops below 500M active parameters, document how in your README.
What counts as "running locally"?
The model must be bundled with or downloadable by your app, and all inference must happen on the participant's own machine. Local runtimes like transformers.js, llama.cpp, ONNX Runtime, TensorFlow.js, Core ML, and TFLite qualify.
Can I use a non-AI API?
Yes. Maps, auth, payments, storage, public data — anything that's not AI inference is fine.
How and when are winners chosen?
Judges review all accepted submissions against the rubric (Impact 25%, Creativity 25%, Technical 30%, Presentation 20%). The top-ranked entry wins. Results and rankings are emailed to all submitters, and the winner is announced in a livestream on Nov 12, 2026.
I can't get my model under 500M — what should I do?
Shrink the problem, not just the model. A 22M model that reliably solves a narrow task beats a 400M model that's unreliable. See the
success guide for advice on model selection.
Who owns my code?
You do. By submitting under an OSI-approved open source license, you retain ownership. See the
terms of use.
