FreshStart

Tracker
Chat interface

Inspiration

The inspiration for our work stems from our experience of seeing our friends and family suffering from addiction. From their experience, we understood two things: 1 - Sobriety apps almost discourage consistent use when a user relapses through harsher words and more blunt actions (clicking “I relapsed”). 2 - Traditional solutions for countering addiction issues include either a person to talk to OR a way to track sobriety. We addressed both of these issues through FreshStart. We are able to infer when users have relapsed through their daily check-ins, during which we gently offer to restart their healing journey. Additionally, we combine options for coping strategies with the tracker itself, creating a central hub for people suffering through addiction.

What it does

FreshStart offers users a way to recover from addiction issues and track sobriety milestones. Through FreshStart, users will check in twice a day, talking about daily affirmations and reflecting on their daily progress. They will also have a personalized AI assistant dedicated to helping them cope with withdrawal symptoms and urges. This assistant also is able to infer when a user might have relapsed and gently reset their journey. Finally, FreshStart tracks how long the user has stayed consistent with their goals and offers a gentle way to reset their journey in the event of a relapse.

How we built it

We divided the application into frontend development, backend development, and AI integration. Frontend(Aryan) was mainly focused on gathering user input, displaying the timer, and gathering responses. The backend(Kedar) handled requests to the AI model and preserving chat history. The AI integration(Rohan) handled prompt-engineering, LLM inference, and sentiment analysis. When we would finish with itemized tasks, we would pivot and help each other until we got to the next stage of our product, making sure to integrate each other's work consistently to mitigate any issues and merge conflicts in future development.

Challenges we ran into

The biggest challenge we ran into was deploying our application to Google Cloud. We had to learn about hosting on the Cloud while managing our Docker backend and communicating throughout the app with the various API calls. While our API was simple, juggling between different specialized chat personas required some ingenuity in how we designed the routes and handled POST requests efficiently. We ran into some issues early on with our LLM backend using the wrong personality for different tasks, but by applying our knowledge in data structures, we were able to resolve this efficiently.

Accomplishments that we're proud of

We are proud to have improved upon the current alternatives in this field and created an app that we know will help people. Through talking to close ones who have suffered before, we learned how to best configure a solution to their needs.

What we learned

We learned how to sideload apps to phones, host servers on the cloud, and work with various frontend frameworks such as Capacitor and Vite.

What's next for FreshStart

We hope to integrate more personalization into the interface, such as displaying pictures of goals or loved ones to remind users what they are fighting for. Additionally, we hope to integrate speech logging for users to give themselves motivating messages and make daily check ins easier.

Built With

Submitted to

TIDALHack 25

Created by

I was the primary frontend developer. I was in charge of creating the user interface for the chats, developing the timer, and storing user states in local storage. I worked with React, Vite, and Capacitor to create the pages and information.

Aryan Bandi
I led the development and integration of the Gemini API model into our app’s interface using LangChain with Python. In addition, I contributed to the frontend by designing and implementing the settings page using React and Capacitor js. This work allowed me to bridge backend AI functionality with a polished, interactive frontend.

Rohan Perumalil
I was the primary backend developer and worked on setting up containerized RESTful API endpoints using FastAPI, Docker, and Python. I also collaborated with Rohan in prompt engineering the Gemini chat model that we used as our urge and check-in manager. In order to ensure our backend doesn't have to be hosted on localhost for our app to work, I deployed our backend container to Google Cloud Run. It was my first time using a hosted cloud service, and I ran into some authentication issues when attempting to push the backend container to Google Cloud that ate up a lot of time trying to solve.

Kedar Panchal