Lusaea - Therapy Companion

Turn therapy insights into real change. Lusaea holds onto your breakthroughs, guides your reflection, and nudges you toward real growth, so therapy meets you in real life, not just in session.

Comment

Inspiration

I started therapy after my best friend lost his fight with depression. While sessions felt helpful in the moment, I wasn’t improving over time — so I quit. A few months later, my dad passed away from the same cause, and I went back to therapy.

I realized the problem wasn’t therapy itself. The insights I gained in session never made it into real life. Everything meaningful stayed in the room.

I later learned I wasn’t alone: 88% of therapy patients feel better immediately after a session, yet only 40% experience long-term recovery. I needed something that stayed with me between sessions, so I could actually practice what I was learning.

What it does

Lusaea is a therapy companion that helps you hold onto your breakthroughs, guides your reflection, and nudges you toward real growth.

During therapy sessions, Lusaea listens in the background and captures key moments. Afterward, those insights are transformed into clear daily steps and reflection prompts, helping therapy continue working in the moments that actually shape behavior.

How we built it

Lusaea was built with React Native and custom Swift UI components for the iOS App Store. The app uses Firebase for backend infrastructure, including Firestore for real-time data storage, Cloud Functions for serverless processing, and Cloud Storage for secure audio file management. The core AI capabilities are all powered by Gemini 3.

Therapy sessions are transcribed using Gemini, with speaker diarization to distinguish between the user and their therapist. From each session, Lusaea extracts summaries, key moments, and therapist recommendations so insights are never lost. Over time, Lusaea builds a longitudinal personal context across sessions — allowing the system to decide which steps matter now and which themes deserve reflection. This bridges the gap between appointments, making therapy more continuous and actionable. All AI outputs are reviewed by a verification agent before being shown to the user, ensuring safety and appropriateness. With each completed session, the user’s personal knowledge bank is updated so future insights become increasingly relevant.

When a user schedules their next appointment, another agent analyzes recent sessions, completed steps, and reflections. It generates a pre-session recap so users walk into therapy feeling prepared, grounded, and on track — rather than starting from scratch.

Lusaea wouldn't be possible without these features from Gemini 3:

  • Audio understanding and speaker diarization
  • Long-context understanding (1M+ token window)
  • Structured outputs
  • Function calling

Challenges we ran into

A key challenge was managing AI context. While richer context improves quality, a single therapy session can reach up to 40,000+ tokens. We designed function-based retrieval so Gemini selectively pulls only the most relevant past sessions, steps, and reflections — carefully budgeting tokens within its long-context window.

Cost optimization was another challenge. We tracked token usage and built an AI abstraction layer to experiment with different models and prompts, balancing price and output quality.

The last challenge was with prompting. Since much of the on-screen text is generated by Gemini, ensuring the proper tone and language was critical for the user's experience. Finding the correct prompts that capture the right brand identity and deliver the best UX required many iterations before finding what worked consistently at scale.

Accomplishments that we're proud of

We’re most proud of publishing Lusaea to the iOS App Store in under a month, marketing it through personal networks and communities, and launching with 30 beta users.

We’re also proud to have worked on a problem with real-world consequences that are deeply personal to us. Lusaea is the first therapy companion designed to be present during therapy sessions — not just something used before or after.

What we learned

The Gemini Hackathon gave us great experience in AI development and mobile app development. We learned how to optimize different models for different tasks and weigh architectural decisions the way enterprises do.

Working with Gemini 3 highlighted how much output quality depends on system design, not just prompts. Clear role definition, instructions, and explicit constraints had a greater impact on reliability than incremental prompt tweaks, emphasizing that Gemini performs best when treated as a component within a well-designed system rather than a single prompt-response tool.

What's next for Lusaea

Next, we plan to expand our pilot, deepen personalization across sessions, and continue shipping new features. We’re also exploring broader platform support (android) and additional tools that help therapy insights translate into lasting change.

Share this project:

Updates