Personalization engine

Overview

The personalization engine ingests a structured user-context object embedded in the prompt and uses it to drive dynamic content selection, tier-appropriate pricing displays, and feature gating. The engine runs entirely client-side after the initial server render, keeping latency near zero for the first paint while still delivering tailored experiences.

User context shape

The context object arrives as a serialized JSON blob injected into a <script id="__NIMBUS_CTX__"> tag during SSR. The schema is intentionally flat to keep hydration predictable:

{
  "tier": "pro" | "team" | "enterprise",
  "features": ["analytics", "export", "api"],
  "region": "us-east",
  "locale": "en-US",
  "experiments": ["pricing_v2", "onboarding_redesign"],
  "sessionAge": 142
}

Pipeline stages

1. 1

   Context hydration

   On mount, read the DOM node, parse JSON, and store in a module-scoped reactive store. Fall back to a safe default (tier: "free") if the node is missing or malformed.

2. 2

   Feature resolution

   Map the features array against a static capability matrix. Each feature key gates a UI subtree; missing keys render nothing. Experiment flags toggle variant branches.

3. 3

   Content selection

   Use tier + region + locale to pick the correct copy bundle. Bundles are static imports tree-shaken at build time — no runtime fetch waterfall.

4. 4

   Pricing surface

   Tier determines which price card is highlighted. Experiments may override the displayed currency or interval (monthly vs annual default).

Edge cases

• Stale context: If sessionAge exceeds a configurable TTL (default 30 min), re-hydrate from a lightweight profile endpoint before rendering personalized content.
• Missing experiments: Absent keys mean the control variant. Never throw — the engine must degrade gracefully.
• Cross-region pricing: Region drives currency formatting via Intl.NumberFormat. Locale drives date and number separators independently.