OpenAI’s first hackathon in Asia, in conjunction with @GovTechSG
Participants applied the OpenAI API to public good use-cases. The winning projects were:
#1 - Automated OneService for SG’s increasing elder population. Using the Realtime API, senior citizens can file municipal service cases in any language.
#2 - Excalibur, a language learning game. Built with Audio in Chat Completions, dall-e, and o1-preview, children can use speech to control a video game character while learning Mandarin.
#3 - Medicine Adherence, built with the Realtime API to keep seniors on track with their medicine (reminding them to take “the yellow pill”) and diagnose side effects.
Hope this will be the first of many in Asia.
Hey! I was one of the developers from the Medication Adherence team- super cool experience to participate in the hackathon, was extremely inspiring all around! Thanks OpenAI for hosting it!
Nice!
We’ve got a great dev community here with a lot of great people. Hope you stick around!
Can you tell us more about your project?
Welcome @Mystichunterz
Echoing @PaulBellow, I’d love to read about your project. feel free to create a new topic about it or reply to this one
Thank you!
Our problem statement was the challenge of Medication Adherence, which is essentially, patients forgetting to take their prescribed medication, especially with Singapore’s aging population.
During the hackathon, we built an interactive medication chatbot using OpenAI’s technologies like the o1-preview reasoning model, GPT-4o, and the Realtime API.
The features included:
- Notifying elderly users when it’s time to take their medication.
- Answering questions about dosage, food requirements, next doses, and potential side effects.
- Generate a curated report for the user’s doctor, highlighting urgent warnings (e.g., side effects) and a general health summary.
Realtime API was used for bi-directional speech-to-speech interaction with GPT-4o, so the user would interact with the chatbot verbally using their microphone, as well as for transcribing the conversation for later use in the report.
To prevent GPT-4o from hallucinating medication details, we fed it the medication data and specifics into its context window. For future iterations, we imagine this functionality would be improved by fetching data from publicly available health APIs to automate the process.
Finally, we used the o1-preview model to create the doctor’s report. We gave it the conversation transcript, along with a provided report template. The model then generated detailed, actionable insights for healthcare providers.
During the demo, we showcased this functionality by sending the report to a provided email address, to simulate how a doctor might receive the report.
Let me know if you have any further questions