Tranquilo is an AI-powered mental health journaling platform developed during the nwPlus 2025 Hackathon at the University of British Columbia. The platform offers users:
- 🧠 A chatbot-like feature for mental health support.
- 📊 A data analytics dashboard for deeper insights into their emotional well-being.
- Interactive Chatbot: Provides context-aware and therapist-like responses to user prompts. There is a voice input and output option as well!
- Mood Tracking Dashboard: Visualizes the user’s mood trends over time for better self-awareness.
- Personalized Insights: Uses Retrieval-Augmented Generation (RAG) to generate responses based on the user's prompt and historical data.
The backend architecture of Tranquilo leverages modern AI techniques and ML fundamentals, designed with scalability and efficiency in mind:
We used the Mental Health Counseling Conversations dataset from Hugging Face, which includes over 3,500 rows of patient and therapist conversations.
- BERTopic was employed to model the raw dataset, identify themes, and generate human-interpretable topic titles using Llama3.2.
- Topics and associated data were stored in a SQL database for efficient querying.
- The trained BERTopic model predicts the topic of a user’s input prompt.
- Therapist responses matching the topic are retrieved and reranked using cosine similarity to determine the top 3 most relevant responses.
- An LLM performs sentiment analysis to evaluate user mood.
- A brief topic summary of the user’s input is generated and stored in a time-series SQL database.
- The user’s prompt, the top 3 therapist responses, and 5 days of user mood history are used to build a Retrieval-Augmented Generation (RAG) pipeline powered by Llama3.2.
- This pipeline generates a therapist-like, context-aware response, integrating current and historical user data.
- Team: Angus Lau, Parth Patel, Akshat Kalra, and Sai Pusuluri.
- Special Challenges: Successfully deployed the entire architecture locally for the sponsor prize.
- Key Achievement: Delivered seamless, meaningful responses through a robust AI pipeline, making mental health support more accessible and personalized.
- Programming Language: Python, TypeScript
- Backend Framework: Flask
- AI Models: BERTopic, Llama3.2 (locally hosted)
- Database: SQLite
- Deployment: Fully locally deployed architecture for data privacy and security. Utilized Ngrok to expose Flask APIs publicly.
- Cloud deployment for scalable use.
- Enhanced analytics and dashboard features.
- Integration with other softwares such as Notion
- Integration with Amazon Echo
A huge thank you to my incredible teammates—Angus Lau, Parth Patel, and Akshat Kalra—for their collaboration and expertise. Special recognition to Akshat for his love of Flask and all the fun moments during development!
This project is licensed under the Apache-2.0 License.
Tranquilo is a hackathon project and is not a substitute for professional mental health advice. Always consult with a licensed therapist for any mental health concerns.