Hey Doc!

Helping doctors make more mindful decisions when working with diverse populations. By creating an anonymous pre-screening system, we hope to bridge the gap between healthcare and culture.

Comment

Hey Doc's Creative Edge & Solution

Diverse populations struggle to find proper care, specifically, these more diverse and underrepresented groups often face up to 30% more misdiagnoses, and often take longer to come to diagnosis. This is where we thought, instead of replacing an existing system, it would be more beneficial to slowly change and educate the current system.

Vision for Expansion

"Making health care more inclusive and equitable." We aim to create an inclusive AI-powered health assistant that can understand, and adapt to, the unique backgrounds of users. We believe there are multiple methods to achieve this, including voice recognition and translation, bias detection, and AI-driven symptom analysis*, we empower users with **personalized and fair* medical insights, reducing disparities in healthcare access. Likewise, we also aim to help wait times which may be lowered due to gained insight from patients.

Challenges Faced

Challenges are nothing but times to learn and prove yourself

  • Finding a good dataset to train the machine learning model and fine tune it.
    • We found multiple datasets that either had repeating or just incorrect data
  • database integration with frontend and backend.
    • Towards the end it took the most amount of time, we were able to connect everything in the end, but that's a lesson learned to not connect until last minute
  • Making a custom machine learning model which is bias-aware. Medical literature itself contains inherent biases.
  • Ensuring accurate symptom translation across multiple languages (future challenge)

What we finished

  • We used a machine learning model to analyze and determine patterns
  • Created a database that allows us to ask information based off a get request
  • Natural Language processing by ChatGPT after ML feeds the weights/words
  • Connected all 3 features through separate API's, a little inefficient, but it works

How AI was used and Incorporated

  • For starters, AI was used throughout the whole process for debugging, however pages and components were a mixture of the two, with the more technical and design aspects of the pages such as handling requests being sent to the database
  • We trained our own Machine Learning model which gives us weights and keywords, which we then put into OpenAI (api) with a system prompt to give us a natural sounding "recommendation" that the doctor can use and keep in mind

Where do we see Hey Doc! In the Future?

  • Many of our team members did actually want to build this out, outside just physicians specifically for this product, but others who seem like they might know more of the root cause.
  • The ML model works like a charm, when ideally we'd be ale to add and overwrite and find a perfect dataset, we believe our results would've been at least 3 times more accurate with a proper dataset, as hallucinations were also present until another training method was used.
  • As a whole, we'd like to create a system that removes the middle man for most Dr appointments. You'd only need when you need to talk to one person
Share this project:

Updates