Software,
shipped honestly.
For teams using software, machine learning, data or hardware to solve a defined real-world problem. We grade the working prototype, the engineering decisions and the ethics — not the founder narrative.
Three deliverables.
Something a juror can run on a phone, a browser or a desk. Demos in slide-decks alone are not accepted at the showcase round. Source code is submitted in a private repository.
Five pages on architecture, dataset provenance, evaluation method, known failure modes and what the team would do next with more time. Teams that pretend to have solved harder problems than they have are marked down.
One page on who could be harmed by the tool if it shipped, how the team would mitigate that, and which decisions a real operator would still have to make. We borrow this requirement from research-ethics review boards.
Engineers, not pitch coaches.
The Digital & AI track is led by Dr Aaron Walsh out of Imperial College Enterprise Lab in London, by Newlab’s engineering bench in New York, and by Cyberport’s senior engineer-in-residence cohort in Hong Kong. Mentors are working practitioners — they read code, not slides.
Teams whose work touches on health, fairness or safety-critical systems are paired with an additional reviewer drawn from the academic council. Our reference points include the Lemelson-MIT InvenTeams programme and the James Dyson Award’s engineering-led judging.
Selected work, 2024–25.
A lightweight triage tool for after-hours under-resourced clinics. Open-sourced; piloted in two community clinics in Sha Tin.
A bus-network simulator for school districts evaluating route changes. Adopted as a planning aid by Lambeth’s schools transport office.
An OCR pipeline for low-resource Chinese palaeography. The team’s evaluation methodology was the standout part of the entry.
Run it, write it, say who could be hurt.
The Digital & AI track is for teams whose intervention is a tool: software, a model, a dataset pipeline, a piece of hardware, or a combination. Process after shortlist: architecture sketch in week 3, a runnable increment by week 6, a frozen demo branch before you fly, and a five-page technical note that covers provenance, evaluation, failure modes and what you would do with more time. Source lives in a private repository shared with the track lead. Slide-only demos are not accepted at showcase.
Examples of work that has cleared the bar: a triage helper trained on a public dataset and checked with two practising nurses, with a note on what the model must not decide; an energy-mix calculator for a named school district using published tariff data; a hardware sensor a facilities manager could mount without a research assistant. Examples that have not: a chatbot wrapped around an API with no evaluation; a “platform for all students”; a computer-vision claim with no confusion matrix and no consent story.
What good looks like in the ethics note is specific. “AI can be biased” is not an ethics note. “If this tool is used to refuse an appointment, the operator still has to see the patient, and here is the interface that forces that” is. Safety-critical and health-adjacent work is double-read by the ethics chair. We would rather you pick a smaller, truer problem than a fashionable one you cannot instrument.
Inline FAQ: Can we use closed-source models? Yes, declare them and do not pretend you trained what you called. Can we open-source after the Final? Yes, and we will help if you ask. Can a Brand idea live on this track because there is an app? Only if the app is the intervention and the jury can run it — otherwise use Brand & Communication.