Dementia patients rely on AI assistants daily. When those assistants go offline mid-conversation, the result isn't just an inconvenience. It's confusion and broken trust for someone who can't troubleshoot an error screen. We built BUDDY Relay Care so that never happens. What it does
It wraps any LLM-backed assistant with an automatic failover engine. When TrueFoundry crashes, it switches to OpenRouter. When OpenRouter is unavailable, it falls back to Ollama, all without losing conversation context.
Getting conversation state to survive a mid-session provider switch cleanly. Each provider has a different API shape, so normalizing the request/response format across TrueFoundry, OpenRouter, and Ollama was a real long and stressful feat
The failover is invisible to the user. You can kill the local model mid-conversation and the assistant keeps responding without missing a beat. Same context, same tone, just a different engine underneath. What we learned
Resilience isn't just retrying failed requests. It's about classifying the failure, choosing the right recovery path, and communicating honestly with the user when something has changed. What's next for BUDDY Relay Care
Expanding to voice-first interfaces via Alexa, adding predictive health monitoring through the YouCam module, and packaging the Relay Orchestrator as a standalone library any FastAPI app can drop in.
Built With
- fastapi
- ollama
- openrouter
- python
- truefoundry
Log in or sign up for Devpost to join the conversation.