TUMAIgotchi | Devpost

Inspiration

TUMAIgotchi was created by CoolGroup to address a familiar problem: sports watches provide endless health data, yet they don’t provide the discipline or motivation needed to act on it. People drown in metrics but rarely feel emotionally connected to them. We realized that humans are often far better at caring for others than for themselves, and we asked what would happen if your wellbeing shaped the life of something you actually care about.

What it does

TUMAIgotchi turns your health data into a living digital companion whose condition reflects your own. Using generative AI, every user receives a unique creature with its own style and personality. It doesn’t merely show your biometric data; it reacts to it in real time. If you’re sleep-deprived, it wakes up tired. If you’re stressed, it becomes uneasy and encourages a moment of calm. If you overtrain, it struggles and asks you to slow down. When you’re active and balanced, it becomes energetic. Instead of charts and scores, you get an emotional, intuitive connection that naturally motivates healthier behavior.

How we built it

We developed TUMAIgotchi using the Oniro Development Environment and ArkTS to integrate with HarmonyOS smartwatches. AWS Bedrock powers our agentic AI system, allowing multiple models to interpret biometric data and generate personalized, proactive companion behavior. A custom generative pipeline creates the companion’s visuals and expressions, while a modular backend ingests and processes health metrics.

Challenges we ran into

Huawei’s strict handling of medical data introduced identification and access hurdles that required creative workarounds. Designing an agent that feels genuinely supportive rather than intrusive demanded careful balancing. Rendering expressive AI-generated visuals within smartwatch constraints proved unexpectedly difficult. Despite these technical and design hurdles, we built a system that feels alive and emotionally engaging.

Accomplishments that we're proud of

We created a modular and scalable architecture for agentic health behavior, a functioning multi-agent reasoning pipeline, and a companion experience that users quickly connected with. The combination of emotional design and real-time biometric interpretation was especially rewarding to see working on-device.

What we learned

We learned how challenging—and exciting—it is to build AI that behaves rather than simply answers. We saw how emotional feedback can be more motivating than numerical data. And we confirmed that bringing a digital creature to life on a smartwatch is harder than it looks, but absolutely worth it.

What's next for CoolGroup

We plan to extend TUMAIgotchi with a richer phone app, deeper personality traits that adapt to long-term habits, and real-time interaction during workouts. We also want to explore social features that let users—and their companions—meet, collaborate, or simply encourage shared activity. Our goal is to grow TUMAIgotchi into a full ecosystem that makes health feel personal, playful, and emotionally meaningful.