Stay tuned for more, this project actively being worked on.
Inspiration
There are two core inspirations to LeetDiagnose. The first one is an entertaining video where medical students have 60 seconds to determine the right diagnosis based on a scenario by asking questions. The second one is LeetCode for it's dominant success among computer scientists and puzzle enjoyers.
What it does
LeetDiagnose let users try to diagnose simulated patients via a series of inquiries, and are graded based on the empathy shown, professionalism, accuracy, and relevance of inquiries. Users are provided with a feedback of their performance.
How we built it
The front-end is made with react, which communicates with our Flask backend server's API endpoints. The Flask server communicates with Firebase for all user data and stored scenarios. It also communicates with Gemini's API to interpret user input.
Challenges we ran into
- Finding the right structure to design the game, ensuring it gave an intuitive and interactive feeling for users whilst minimizing the inconsistencies and reminders of LLMs' flaws.
- Server-side issues, specifically connecting the front end to the back end.
- Security problems that we had to reinforce using NGINX.
- Deploying
- Finding an appropriate data set to write scenarios with.
- Uncooperative UI
Accomplishments that we're proud of
- Interactive and responsive user interface.
- Smooth user experience.
- Ensuring accurate and consistent LLM usage.
- Beautiful UI.
What we learned
We gained experience in integrating front-end and back-end technologies, using AI for dynamic game interactions, and improving security measures.
What's next for LeetDiagnose
- More scenarios.
- Improvements to the interactions and feedback, based on user feedback.
- Credited users will be able to upload their own scenarios.
- Integrate a social aspect to the app.
Log in or sign up for Devpost to join the conversation.