probe_query_relevance execution logic — DEJAN Codebase Explorer

The relevance-probe engine iterates over active entities or queries against the selected models, formats each call from a metaprompt template, dispatches them across concurrent worker queues, and aggregates the binary yes/no responses in the database to calculate relevance percentages.

Withheld — proprietary.

sequenceDiagram
    autonumber
    actor User
    participant Route as API Route
    participant Probe as Relevance Engine
    participant DB as Database
    participant LLM as Provider APIs

    User->>Route: Start relevance probe (brand, samples)
    Route->>Probe: Run in worker thread
    Probe->>DB: Fetch active models, items & metaprompt
    Probe->>DB: Open run (status running)
    loop For each item x model x sample
        Probe->>LLM: call(model, prompt)
        LLM-->>Probe: Result ("yes" / "no")
        Probe->>DB: Record response
        Probe->>DB: Update progress
    end
    Probe->>DB: Close run (status complete)
    Probe-->>Route: Return summary counts

Referenced by