Inspiration

While thinking of issues we came across a news website mentioning that students in some universities around the UK are waiting weeks just to be able to join mental health groups or consulting sessions. Students around the world are stressed, whether with classes, sleep, maintaining a social life or personal issues they're facing. We thought we could make a platform to let them be heard and not be judged.

What it does

Once you're in a bubble, depending on what issues you're facing, an AI will lead the session the way a trained counsellor would. Keeping the conversation structured, making sure everyone feels heard.

How we built it

For planning we used Manus to research the current student mental health landscape. We built the solution using Z.AI. We benchmarked multiple AI models and found DeepSeek V4 to perform the best in support group situations recalling and interacting with multiple people.

Challenges we ran into

Biggest challenge was making sure the AI felt like friendly and inviting rather than very AI like. We didn't want students to feel like they didn't want to talk and share because they were talking to an AI, so we wanted to make sure the AI doesn't give that vibe. We analyzed support group conversations and AB tested across them to make our solution personalized to each situation.

Accomplishments that we're proud of

The home page animation was one of the things we wanted to implement from the very start as it feels like you're entering a bubble. Additionally, when we tested conversations with the AI they felt really helpful, and we truly see this product being something that students would use.

What we learned

Designing for emotional safety is a different problem compared to designing for usability. We had to make important decisions such as the colour choice, the home page and how interactive it is. We didn't want students to click and immediately regret it and change their mind for joining, which is why we made sure everything feels inviting and feels safe.

What's next for Circle

A safety layer that not only implements guardrails for the AI, but for the human layer so there is an added layer of emotional intelligence to flag when an AI needs to handoff to a human in extreme situations.

Built With

Share this project:

Updates