Preferences & affinities

Understand what the user repeatedly gravitates toward.

Preferences and affinities capture recurring interests in food, places, styles, activities, and experiences. Rollup infers them across distinct moments rather than treating every photo as a separate vote.

{
  "concept": "specialty_coffee",
  "strength": 0.84,
  "confidence": 0.89,
  "supporting_moments": 36
}

Strength and confidence#

Strength describes how pronounced an affinity appears. Confidence describes how well the evidence supports that inference. Both are illustrated on a 0–1 scale, but neither is a guarantee of what a user will choose next.

Domains narrow a read#

const profile = await rollup.profile({
  domains: ["food", "travel"]
})

Domains help a feature request relevant context. The underlying abstraction is the user’s persistent understanding; applications do not need to choose a domain to use profile().

Personalize with room for correction#

Use affinities to rank or suggest, and let explicit user choices take precedence. A sparse or absent preference should leave room for discovery.