LATEST NEWS

Registration for the Medprompt Hackathon 2025 is now closed.
Thanks to everyone who registered! Please complete your prompt submissions directly on Kaggle before the deadline. If you encounter any issues or haven’t received the competition access details, check your spam or junk folder, or reach out to us for assistance.
The Medprompt Hackathon 2025 competition is now officially open!
Competition links have been sent to all registered participants via email. If you have not received the competition link, please check your spam or junk folder, or contact us for assistance.
Thank you to everyone who joined the Medprompt Hackathon 2025 launch event.
The launch event has now concluded.
The recording and materials used during the event are now available here.
Due to unforeseen circumstances, the Medprompt Hackathon 2025 launch event originally scheduled for August 19, 2025, at V322 (Lecture Theatre), Jockey Club Innovation Tower, Hong Kong Polytechnic University will now be held online only.
We apologize for the short notice and any inconvenience caused. We have sent emails to registered participants with the online meeting link details. For those unregistered and interested, you can click here to join directly at the scheduled time (18:00, Hong Kong SAR Time).
Please contact us if you have any issues when joining. Thank you for your participation!
Medprompt Hackathon 2025 is now accepting applications. The deadline to apply is 15 September 2025. To join, click the “Sign Up Now” button at the top of the page and be part of the future of AI-driven healthcare!

Overview

While large language models can deliver impressive results, they often demand substantial computing resources—posing barriers in terms of cost, accessibility, and environmental impact. Compact models enable secure, on-site diagnosis support, minimizing the need to transmit sensitive patient data to external servers, which are more practical for real-world use such as healthcare.

However, these smaller models are particularly reliant on the clarity and effectiveness of their prompts to perform well. Thus, creating a single, cost-efficient “super” prompt—an instruction that enables resource-limited models to approach expert-level diagnostic performance—can make advanced AI tools more accessible in clinical settings without compromising privacy.

Objective

In this competition, you’ll be challenged to create a single, highly effective prompt for small language models (under 70B parameters) that enables them to generate accurate, one-sentence medical diagnoses from a diverse set of >200 clinical case reports.

Rather than building a new AI model, your task is to design a prompt that can consistently guide existing, resource-constrained models to extract and summarize complex clinical narratives with precision.

By advancing prompt engineering, this challenge aims to help models make accurate diagnoses from complex clinical cases—contributing to safer, more efficient, and privacy-preserving AI tools for healthcare professionals.