The integration of Gemini 3 Pro in Routines is the engine of its “SmartDigital Coach” architecture, transforming the app into a proactive, context-aware stretching assistant. By leveraging its advanced reasoning and native Function Calling, the application autonomously orchestrates personalized workouts based on real-time environmental and biological data.
Key features utilized include:
Agentic Tool Use via LangGraph: Gemini 3 Pro serves as the decision-making core within a StateGraph. It intelligently triggers specific tools—like the Strava API to assess fatigue or the Google Calendar API to find training gaps—ensuring routines are both physically optimal and logistically feasible.
Multimodal Context Ingestion: The model synthesizes diverse data streams, including local weather and granular user profiles. This allows for dynamic adaptations, such as pivoting to indoor mobility exercises during rain or suggesting specific recovery stretches after high-intensity activities detected via Strava.
Structured JSON Outputs: Gemini’s precise instruction following guarantees that complex routine data is delivered in a strict JSON format. This enables seamless synchronization with the PWA frontend, providing a robust and reliable user experience.
By moving beyond static templates, Routines uses Gemini 3 Pro to deliver a truly intelligent fitness journey that understands the “when” and “why” of every movement.
Built With
- gemini
- langgraph
- next
- node.js
- pwa
- typescript
Log in or sign up for Devpost to join the conversation.