Published before registration

Rules and scoring

The card, the rounds and the tie-breaks are all public in advance, so a result can be explained to the team that came fourth.

The scoring card

One hundred points. The same card in the first round and in the final.

PointsCriterionWhat the judges assess
25Value for the userA specific problem and a clear outcome from solving it.
30Prototype and use of AIA working core scenario, and what AI actually contributes to the result.
20Quality testingTests, examples of failures, and comparison with the current approach.
15FeasibilityData requirements, running costs and a clear next step.
10OriginalityA meaningful difference from common solutions.

How winners are selected

01

The first round is read by AI

Your first judges will not be human. GPT and Claude read every project against the card above — the same hundred points for every team — and their two scores are averaged. This runs live on the OMNI AI Summit stage at 11:30 on day two.

02

30 finalists

The 30 highest averages go through, and all 30 take prize places. One ranking across every project, rather than a quota per room, keeps the bar identical for everyone.

03

The final is judged by people

From 15:00 the finalists pitch on stage to an international jury and a full hall. The jury scores them afresh on the same card and the result is the average. Ties go to the prototype score, then to quality testing.

What you submit

All of it is due at 21:00 on day one, and the first round reads exactly what arrived by then.

  • A pitch deck

    The deliverable the first round reads. Due at 21:00 on day one.

  • The user and the problem

    Who this is for and what goes wrong for them today.

  • A demo link, if you have one

    Optional. Something a judge can open — if it needs setup, say exactly how.

  • A GitHub or GitLab link

    The repository with your code. It need not be public, but judges must have access to verify.

  • A video of up to two minutes

    The core scenario, running.

  • Models, data and components used

    Disclosure is mandatory, including libraries and public templates.

  • Quality testing results

    Your tests, your failures, and a comparison with how the task is done today.

The rules in full

  1. 01At least one and at most five people per team by the start of development.
  2. 02Research of ideas, tools and open data beforehand is allowed. The project itself must be built after the hackathon starts.
  3. 03Existing libraries and public templates are allowed if their use is disclosed. Previously developed products are not eligible.
  4. 04Every submission discloses the models, data and existing components it used, together with quality testing results.
  5. 05Build time ends on day one. The pitch deck is due at 21:00 and the competition version cannot change after that; teams may still prepare how they present it on stage.
  6. 06Teams keep the rights to their projects. A public repository is optional, but judges must have access for verification.
  7. 07The first round is read by GPT and Claude on the published card; the 30 highest scores go to the final. Jurors in the final declare conflicts of interest and do not assess teams they are connected to.

Project ownership

Teams keep the rights to their projects. Publication of materials by the organiser, and any pilot with a partner, require separate agreement. Apply if that works for you.