Prove It—Building Evidence for Trustworthy Multimodal AI Deployment

Hackathon registration is now open — 15 May 2026.

If you are registered for the MultimodalAI'26 workshop, you are already eligible. Please complete the hackathon registration form to secure your place.

Registration closes 4 June 2026. Full participant details — strand guides and test notebooks — will be sent to all registered attendees on 4 June 2026. Watch your inbox.

The Challenge

AI systems are being deployed in ICUs, autonomous vehicles, social housing, and robotic platforms — but can the people who depend on them actually trust them? Trust has to be earned with evidence, not assumed.

The MultimodalAI'26 Hackathon is a deployment-centric event. It invites registered workshop attendees to spend a day doing what most AI development skips: rigorously auditing whether AI systems that combine multiple data types are safe, fair, and honest about their own failures.

No single “best model” wins here. The strongest entry finds the most important real-world failure mode, explains it clearly enough for a practitioner to act on, and produces a structured report that could actually inform a deployment decision.

Most AI hackathons ask you to build the best model. This one asks you to prove it deserves to be trusted.

Working in teams of 2–4, you will spend a day evaluating AI systems across four consequential real-world scenarios — using synthetic datasets carefully designed to reflect genuine deployment conditions — and producing structured evidence of whether it should be trusted. You will not be judged on accuracy alone. You will be judged on evidence.

Four Strands. Four Questions.

Each strand provides guided notebooks, synthetic datasets, and parallel track roles so the work can be divided across your team. You do not need to be an AI researcher to participate — domain knowledge is as valuable as code.

Why Join?

• Strong submissions may be invited for publication outputs.
• Our previous sprint produced a perspective paper in Nature Machine Intelligence.
• Strong deployment-centric findings from this hackathon will be considered for OMAIB benchmark contributions and future publications.
• The EU AI Act creates demand for auditors. Spend a day becoming one.
• Have fun building something that funders and collaborators can understand and use.
• Good food, good teams, real contribution.

Awards

The winning team will receive an Amazon voucher. Two teams will be recognised at the MultimodalAI'26 Workshop, taking place on 11–12 June 2026:

• 🥇 OMAIB Future Path Award — for the submission most likely to become a lasting benchmark contribution.
• 💡 Best Multimodal Idea — for the most insightful finding across modalities.
• 🖥️ Best Demo/User Experience — for the most engaging and usable demonstration.

Winners are announced on Day 1 of the workshop (11 June); the prize is awarded on Day 2 (12 June).

Practical Information

• Laptop: Participants are expected to bring their own laptop with Python 3.11+ installed.
• Internet: Eduroam is available at the venue. Guest Wi-Fi details will also be provided on the day.
• Food: Lunch and refreshments will be provided.
• Further details: Environment setup instructions and any additional requirements will be included in the participant briefing sent on 4 June 2026.

Key Facts

Stay Informed

Full strand details and test notebook with detailed instructions will be sent directly to registered MultimodalAI'26 workshop & hackathon attendees on 4 June 2026.

If you have a question in the meantime, contact the organising team: multimodalai26-group@sheffield.ac.uk