Preference Inference

Infer a user's attraction preferences, lifestyle traits, and short bio from their photos and (optionally) their existing self-written bio.

POST /v1/profile/infer-preferences

Request

{
  "user_id": "usr_abc",
  "photos": [
    "https://cdn.your-app.com/photos/usr_abc/1.jpg",
    "https://cdn.your-app.com/photos/usr_abc/2.jpg",
    "https://cdn.your-app.com/photos/usr_abc/3.jpg"
  ],
  "bio": "optional existing bio text",
  "name": "optional user display name",
  "age": 28
}

Field	Type	Required	Notes
user_id	string	yes	Your stable identifier; opaque to us
photos	string[]	yes	1–5 publicly-accessible HTTPS URLs
bio	string	no	Up to 2,000 characters
name	string	no	First name only is fine
age	number	no	Used to weight peer-context in inference

Response

{
  "request_id": "req_01H…",
  "user_id": "usr_abc",
  "preferences": {
    "lifestyle": ["outdoorsy", "creative", "metropolitan"],
    "personality_traits_sought": ["curious", "expressive"],
    "interests": ["live music", "hiking", "indie film"]
  },
  "bio": "Outdoorsy creative who'd rather be at the trailhead than the office. Wants someone curious enough to show me their favorite indie film and patient enough to enjoy the slow drive there.",
  "gender": "female"
}

Field	Type	Notes
preferences	object	Three short arrays — lifestyle, traits-sought, interests
bio	string	Polished bio, 1-2 sentences, ready for direct display
gender	string	female, male, or unknown if confidence < 0.8

Latency

P50 ~ 4–6 seconds, P95 ~ 8–12 seconds. The endpoint runs vision analysis on each photo plus a final synthesis call, so latency scales with photo count.

For a real-time UX, call this from a background worker after onboarding completes rather than blocking the user on it.

Common errors

Code	Why
invalid_photos	Fewer than 1 or more than 5 URLs, or a URL returned non-2xx
unsafe_content	One or more photos failed our content safety filter
quota_exceeded	Free tier monthly cap reached — see Pricing