Relevance Studio Overview

Relevance Studio is the admin-only control room for tuning, observing, and scaling AACsearch. It is shipped as a set of 16 panels inside apps/saas and is intentionally separate from the day-to-day Dashboard: Studio is where a search engineer or platform owner tunes ranking, runs LTR experiments, inspects cluster health, manages cross-region replication, and debugs the read path.

If you are a content editor adjusting synonyms or curations, use the Dashboard → Relevance Tuning page instead. Studio is the heavier surface — it requires admin role and exposes features that can change ranking for the entire organization.

Access

• Role gate: Studio requires admin or owner on the active organization. Member-role users see a 403 banner instead of the panel.
• Plan gate: the LTR and Cross-region areas additionally require the Scale plan or above (see Plans & Limits).
• Route: /[orgSlug]/studio/<area>/<panel>.

The Studio surface tree

The 16 panels are grouped into 5 areas. The tree below is the same shape that the sidebar renders inside the SaaS app.

flowchart LR
    classDef area fill:#dbeafe,stroke:#1d4ed8,color:#1e3a8a
    classDef panel fill:#fef3c7,stroke:#b45309,color:#78350f
    classDef root fill:#ede9fe,stroke:#6d28d9,color:#4c1d95

    Root["Relevance Studio<br/>(admin-only)"]:::root

    Root --> Relevance["1 · Relevance"]:::area
    Root --> LTR["2 · Learning to Rank"]:::area
    Root --> Health["3 · Health & Scale"]:::area
    Root --> CrossRegion["4 · Cross-region"]:::area
    Root --> Analytics["5 · Analytics & Debug"]:::area

    Relevance --> R1["Field weights"]:::panel
    Relevance --> R2["Boosts & decay"]:::panel
    Relevance --> R3["Presets"]:::panel
    Relevance --> R4["Personalization"]:::panel

    LTR --> L1["Click feedback"]:::panel
    LTR --> L2["Training runs"]:::panel
    LTR --> L3["Models"]:::panel
    LTR --> L4["A/B tests"]:::panel

    Health --> H1["Cluster health"]:::panel
    Health --> H2["Capacity planner"]:::panel
    Health --> H3["Slow-query log"]:::panel

    CrossRegion --> C1["Replica map"]:::panel
    CrossRegion --> C2["Lag dashboard"]:::panel

    Analytics --> A1["Query inspector"]:::panel
    Analytics --> A2["Ranking explainer"]:::panel
    Analytics --> A3["Audit replay"]:::panel

The 5 areas at a glance

1 · Relevance

Static, editor-driven knobs that shape ranking before any learned model is applied.

2 · Learning to Rank (LTR)

The full feedback-driven pipeline: ingest clicks → debias → train → A/B test → activate.

The full LTR workflow has its own page — see Learning to Rank.

3 · Health & Scale

Operational visibility into the underlying Typesense cluster and the ingest pipeline.

4 · Cross-region

Multi-region replication and failover surface — Scale plan only.

5 · Analytics & Debug

Per-query forensics and reproducibility tools.

When to use Studio

Use Studio when you need to answer one of these questions:

• "Why did this product fall to position 12 last Tuesday?" → Ranking explainer
• "Are clicks improving since we shipped the new title weights?" → A/B tests + Click feedback
• "How much headroom do we have before we need a bigger node?" → Capacity planner
• "Is the EU replica lagging?" → Lag dashboard
• "Why is this 'electric guitar' query slow?" → Slow-query log + Query inspector
• "Should we personalize results for logged-in users?" → Personalization

For pure content-editor work (add a synonym, pin a product) use the Dashboard instead — it stays available to all seats including the member role.

Read more

• Personalization — profiles, segments, session reranking, recommendations.
• Learning to Rank — click feedback, training, A/B tests, activation.
• Architecture → Analytics feedback loop — the underlying event pipeline Studio reads from.
• Dashboard → Relevance Tuning — the editor-facing counterpart.