NeuroDx AI

Inspiration

After researching into mental health we had learned that hundreds of thousands of people die per year for mental health related problems, influencing us to come up with a solution to help these people.

What it does

Our application prompts a user to write a daily journal stored in a local database for security. Through these journal entries, we prompt an LLM in the background which is encrypted via Fernet encryption—an industry standard in applications such as Gemini. Furthermore, it uses a local LLM in the backend to ensure that no data leaks can happen, which has been a recurring issue with cloud APIs like OpenAI recently, such as the leaks in November 2025. The model itself evaluates the daily journal updates and provides feedback after which is then updates the chances of mental burnout among other things. It also has a chat application using an local llm model, the context it's given is the scores from the evaluation of the daily journal throughout the lifetime of usage of the application.

How we built it

We built it with fastapi as the backend and react in the frontend. In the frontend we used package such as dexie which is used for the local database along with fueron encryption. In the backend we used fastapi for the api routes, with ollama to run the local llm used which is gpt-oss:20b, again with fueron for encryption.

Challenges we ran into

We ran into many problems, the main ones however consist of the prompt engineering our application relies heavily in the fact that the models responses are consist and plausabile trusting that it will not fall for noise, along with other things promote bad behaviors. These constraints caused massive tuning of prompt throughout the hackathon, another major problem we ran into was the color scheme for the frontend, the colorscheme is vital in order for the application to be vibrant and seem professional causign us to create a figma mock up along with lots of iterations of color combinations.

Accomplishments that we're proud of

We are proud of our product that we were able to build in just 12 hours. We are proud to make the local database work allowing for a safer and more secure user experience along with the local llm responses which were tuned to match our demographics being high schoolers.

What we learned

We learned a lot, we learned about dexie which was new to all of use and took a while to debug and execute along with the best processes for prompt engineering which was needed for this project