Dear AI-ry

Homepage of Dear AI-ry. Various prompts to interact with, and
Onboarding Email sent by Dear AI-ry, with a personalized message and links to the Web App
Email of a Reminder to Journal, Sent Daily by Dear AI-ry, but can be changed in User Settings
Chat function with Dear AI-ry companion, with excellent semantic memory of chat history

Inspiration

We were inspired by a desire to implement a journalling companion as a way to monitor mental health and promote wellness habits in our communities. Millions of Canadians struggle with all sorts of personal problems without knowing different outlets to destress. This leads to worsening wellbeing and outlook on life, even to the point of developing serious mental conditions such as anxiety and depression. From personal experience, we found journalling to be an easy and effective way to express pent-up feelings and practice mindfulness. However, traditional journalling tools are passive and do not actively help users reflect on their thoughts. In addition to this, we were able to leverage the findings of an expert motivational interviewing study to significantly enhance the benefits of the journalling.

What it does

Dear AI-ry is an AI journaling tool and mental health assistant. Based on the users entry for the day, as well as semantically-similar past journal entries, Dear AI-ry generates key insights into mood, delivers suggested coping actions, and generates a motivational interviewing-inspired question that allows the user to investigate their emotional state and uncover potential solutions.

Daily email notifications are sent by the agent containing these insights and questions. From these emails, users can access a chat and initiate a conversation with the chat to further pursue the line of questioning initiated by the motivational-interviewing inspired question.

How we built it

Our AI agent was built with OpenAI's gpt-oss-120b, using Backboard.io for memory and context-retrieval, and Firebase for database storage, retrieval, and user authentication. We used OpenAI's text-embedding-3-small API to store user's journal entries as vector embeddings in our database, enabling semantic search and similarity comparison.

This allowed us to find previous journal entries most relevant to today's journal entry, giving the model context from days with similar emotions to inform the system’s reasoning process before generating a response. The user is then prompted with the option to view past journal entries with high similarity scores, to reflect and gain additional insight into how they've dealt with these issues in the past.

Backend logic was written in Python, using the Fast API, while the frontend was built with React and Node.js. We created user authentication with Firebase Authentication using email + password and Google. Sendgrid's email notification API was used to manage our alerts to users.

Challenges we ran into

Ensuring proper integration between the frontend and backend was a challenge, since we developed them in parallel. Successfully defining the project structure at the beginning allowed us to avoid a huge headache, but we ran into issues with communication between Firebase, Backboard, our AI agent, and our React frontend.

Account management, scheduling, user authentication, and inbox management were a challenge with our email service. We had to learn a lot on the fly to avoid critical errors in the system, and ensure we had a robust email service in place before deployment.

Accomplishments that we're proud of

Successfully building a fully user-friendly, working MVP within the window alloted to us.
Addressing a problem that all of us have had real personal experience with
Deploying our project before the deadline
Successfully integrating AI agents, vector embeddings, and memory-context tools to build a personalized, smooth user experience

What we learned

We learned about the principles of motivational-interviewing questions, vector embeddings and semantic search, and juggling multiple databases. Additionally, we got exposure to creating semi-autonomous AI agents that personalize user mental health treatment, while maintaining context to ensure a smooth user experience.

What's next for Dear AI-ry

Consulting and collaborating with professional psychologists, CAMH, and leading figures in the motivational interviewing field to refine and improve user mental health treatment.