Context Pipelines
Context pipelines move users from raw signals to an initialized taste state without requiring them to manually curate every surface themselves.
Applied Pattern
A strong external pattern is onboarding from captures. Users submit capture IDs, Mosaic initializes taste clusters, and your application then dispatches asynchronous auto-window generation so the user lands in a non-empty personalized state.
- Collect inputs such as captures or other profile signals
- Call a modeling service to derive structure, such as taste clusters
- Persist session and trace context so the pipeline remains inspectable
- Dispatch async work for downstream windows and derived surfaces
Why Async Matters
Expensive persona-building work should usually happen as background orchestration. That keeps the interaction responsive while still letting the system generate richer Level 2 artifacts moments later.
Signals That Fit This Layer
- Capture uploads and photo-library selections
- Onboarding questionnaires or seed picks
- Imported cultural or social preference sources
- Sentiment activity that is strong enough to justify re-seeding windows
Keep It Light
The clearest applied example is capture-driven onboarding and auto-window generation. Broader social ingestion can fit this layer too, but unless you have a concrete integration already, it is better to frame those as extensions of the same pattern rather than as fully specified APIs.