Inspiration
The point of relieving pressure on the medical industry, whilst ensuring that an incorrect diagnosis that could be fatal would be avoided, led to the idea of creating a system using a technique that I apply to other systems to ensure that hallucinations can be eliminated, medical research can be carried out alongside referrals and patient intake; and all of the patient information recorded is secure, organised and well presented.
What it does
The quorum of Agents is a collection of AI-Agents that work together to use weighting and a vast amount of medical data/research to assist in speeding up and streamlining patient intake, meaning that medical staff have access to on-demand and up-to-date research at the same time as being able to access vitals, symptoms and images to provide decision support and even present them with the most likely ranked diagnosis' to which they are able use to facilitate swift and accurate treatment.
How we built it
The full stack was designed using Anthropic's Claude and several models provided by Nvidia(NIM), AWS' DynamoDB and Vercel for the frontend and backend. And features a Trilingual input system that can process English, Spanish and Chinese. The Medical research was sourced from Pubmed(NCBI), NICE, MedlinePlus, DermNet and Wikimedia, etc. We have included 11339 verified conditions by name and cover everything from common conditions to rare conditions, alongside images and source-cited information anchored to the authoritative guidelines of NICE, ESC, IDSA/ATS, GOLD, KDIGO and the WHO.
Accomplishments that we're proud of
That it started from an idea and it turned into a full-stack and everything works. Mainly, that it's shippable, scalable and ready-to-go.
What we learned
The smooth and quite capable flexibility of AWS coupled with the power of Vercel can lead to a truly stunning product. I had no idea that you could creat and connect a widget to a clients website and relay that information back into the database.
What's next for AI Medical Diagnosis System - Dusky
In-house trained models and more research.
Log in or sign up for Devpost to join the conversation.