Report Generation (SQS Consumer → AI Analysis → Report Storage)

Worker generates an interview report from snapshot and transcript, or marks insufficient evidence without LLM.

Sequence

sequenceDiagram
  autonumber
  participant Settle as settlement.service
  participant SQS as SQS report-generate
  participant Lambda as report-generate handler
  participant Gen as report-generation.service
  participant OpenAI as OpenAI API
  participant Mongo as MongoDB

  Settle->>SQS: publishReportGenerate
  SQS->>Lambda: message
  Lambda->>Gen: generateReport
  Gen->>Mongo: claim report PROCESSING
  alt too short or few answers
    Gen->>Mongo: insufficient_evidence COMPLETED
  else AI path
    Gen->>OpenAI: chat completions json_schema
    OpenAI-->>Gen: structured report
    Gen->>Mongo: COMPLETED report fields
  end

Steps

1. Settlement enqueues { interviewId, organizationId } after first successful credit settlement.
2. Lambda loads the interview and atomically claims the report document as processing.
3. Very short interviews or too few answers get a completed insufficient-evidence report without LLM.
4. Otherwise structured OpenAI output becomes scores, recommendation, and skill assessments.
5. Recruiters poll or fetch report status and content when generation completes.