Personalization & Recommendation Platform

Browse the synthetic campsite catalog as a signed-in guest. The "Recommended for you" rail on the catalog and the "You might also like" rail on a campsite detail page update live as you view and save sites. With no history yet they fall back to a clearly-labelled popularity cold-start. Each pick carries a deterministic "Why this?" chip, and "How these are picked" opens the inspector with the full ranking trace.

Some campsites are flagged fully booked. Open one and use "Notify me with similar campsites" - the engine anchors on that site, applies the same business rules, and returns exactly three similar options via a simulated email (a real send is an owner opt-in, off by default). The recovery is logged in the console.

Open the inspector to see the ranking trace on the uniform trace-row contract - the candidate set, each signal's contribution to the blend, the business-rule filter, the threshold and the diversity pass - all at $0. The single optional metered call is the natural-language explanation; a Cloud/OSS toggle and a cost meter cover it, and it fails closed to the $0 OSS path at the budget cap. The deterministic "Why this?" chips carry the specifics regardless. Everything is labelled synthetic and simulated.

The Console is one unified, role-gated operator surface. Operators see live metrics, the campaign audience (per-customer recommendations with exclude toggles) and the recovery log; a viewer is read-only. Owners additionally tune the ranking weights, the business rules, the thresholds and the email template - every change is audited and re-ranks the engine immediately. The project-page walkthrough video narrates the full tour.