Tranquilo

• 

  Tranquilo-logo

• 

  Project-Architecture

Inspiration

Mental health is a never ending issue and is extremely prominent within students. One popular technique to help control your thoughts is to journal, but not only is constant journalling difficult when you're feeling down, but sometimes we want answers to our entries as well, thus Tranquilo was built.

What it does

Tranquilo records how you feel through voice recognition, and it uses LLM and ML models to provide useful advice.

How we built it

We had a dataset which contained a bunch of patient and therapist interactions. With this we trained a BERTopic model and used it to predict incoming user prompts. Then using advanced RAG and storing our data in a SQL database. The topic modelled user prompt is queried on the database to get all the therapist response for that topic and this vector is matched for similarity using cosine similarity and give out the top 3 similar responses compared with the user prompt in a custom retrieval pipeline and augmented onto a base prompt to generate a relevant response.

Challenges we ran into

• Having same dependencies on all our computers and didn't know how to containerize properly
• Using all LLM's on a local machine
• Picking up SQL with Zero Experience
• Using Eleven labs TextToSpeech Api was easy but setting up the dependencies to make the audio play was challenging

Accomplishments that we're proud of

• RUNNING ALL THE LLM's LOCALLY!
• Learning and successfully implementing advanced RAG
• Got to LEARN SOOO much about LLM models and combining front end with the huge back-end

What we learned

• how to team work amongst teammates
• how to delegate tasks mindfully and coordinate with each other with the results and challenges we faced, also combining backend and frontend
• effectively host a locally run LLM model and what things to improve upon next time

What's next for Tranquilo

• We want to implement a mobile application to increase accessibility for users.
• We want to add in multiple language support for voice recognition.
• We want to add in user authentication using Firebase.
• We want to create a bridge between Notion and Tranquilo to automatically push any journal entry updates in a users notion page to our system.

Built With

• bertopic
• colab
• flask
• javascript
• next.js
• ngrok
• node.js
• ollama
• python
• rag
• shadcn
• sqlite
• tailwindcss
• typescript