Inspiration
The mental health landscape has a fundamental problem: traditional therapy is expensive, hard to schedule, and one-size-fits-all, while existing meditation apps offer only pre-recorded content. This gap inspired me to create Wellness AI - combining AI therapy with truly personalized guided meditations.
What it does
Context-Aware AI Therapy: The app conducts natural text-based therapy sessions, remembering conversations across sessions to build a comprehensive understanding of the user. After each session, it provides personalized insights and actionable takeaways.
On-the-fly Meditation Generation: The most unique technical feature is the ability to generate completely personalized guided meditations. These aren't pre-recorded - they're created in real-time based on:
- The user's therapy conversations
- Specific user requests (e.g., "sleep meditation" or "anxiety before interview")
- User preferences for voice, background audio, and duration
How I built it
Front-end is swift + kotlin on top of kotlin multiplatform.
For the backend, the code is python based, leveraging a variety of LLM and TTS systems. There's a huge amount of orchestration of various LLM calls for maintaining memory within and across sessions, and across modalities.
Challenges I ran into
Creating natural meditation cadence: Unlike normal text generation, meditations require specific pacing, pauses for breathing, and a gentle flow. Significant prompt engineering and post-processing was needed to achieve this.
Context management: Maintaining relevant user history across sessions without overwhelming the AI models required building a lot of data structures. And then there's about 15-20 different LLM calls to create everything within the app (from therapy assistant responses, to suggested meditation topics, to user push notifications).
Accomplishments that I'm proud of
Creating the first app to generate truly personalized guided meditations on-the-fly - something that didn't exist before. And creating a mobile app from beginning to end (with some help), which is something I've never done before.
What's next for Wellness AI
I'm starting to market the app and trying to onboard more real-world users. I did lots of testing, but am awaiting more user feedback before figuring out the next features that I want to build out.
Try it out:
Download now to enjoy an extended free trial:
