Inspiration

A family member passed away of a misdiagnosed illness. This could have easily been caught if they would have sought out a second opinion sooner. I built a tool that encourages patients to make audio recordings of their medical appointments so that they can leverage AI to get a transcript of their healthcare appointments and an automatic second opinion. This is not about using AI to get medical advice -- the point is to show how useful AI can be and motivate people to capture critical information that would otherwise get lost.

What it does

advocate.ai is a patient-directed health tool that helps patients prepare for, record, and review their appointments. The prepare module is a simple interface that helps patients come up with specific questions so that they can make the most out of their time with their healthcare providers. Then, when they do go to a medical appointment, patients are encouraged to ask their providers if they can make an audio recording. Instead of just sharing this full audio file with caregivers and storing the file on their phone or computer, users can upload the audio to a secure account and get an automatic transcription, summary, and critical feedback. They can then ask questions specific to the appointment and access related resources. The site provides an automatic meta-analysis of each appointment and gives patients a heads up about when it might be helpful for them to seek out a second opinion.

How we built it

Started with a simple dashboard. Built out the speech to text diarization feature, then sent the appointment transcript into another AI model for summary and feedback. Used another AI model to enable the user to "chat" with their appointment for further insight. Continuing to add new features!

Challenges we ran into

It was tricky chaining all the required AI calls together and figuring out the right model for each part of the app.

Accomplishments that we're proud of

This is actually incredibly useful! Very happy to have built something with AI that unlocks tremendous real value for end users in a way that is extremely low-cost and accessible.

What we learned

Working with AI gets much more interesting when you combine models and create multi-step calls. You can use an AI to analyze and reflect on the output of another AI!

What's next for advocate.ai

Leverage multi-lingual capacities of AI models to make this patient advocate tool available to users who do not speak English as their primary language.

Built With

  • exa
  • gemini
  • next.js
  • next.js-with-typescript-on-the-frontent.-ai-model-inference-provided-by-google-ai-studio
  • openai
  • react
  • replicate
  • supabase
  • typescript
Share this project:

Updates