Human-in-the-Loop
Human-in-the-Loop (HITL) review ensures that AI outputs are validated by humans before becoming part of Squad’s learned knowledge. This is critical for high-stakes environments where accuracy and oversight are non-negotiable.
What to Expect
- Using the Review Panel during chat sessions
- Reviewing pending items: understanding the AI’s reasoning and results
- Approving responses (promotes the query to approved status)
- Rejecting responses with reasons
- Amending responses: correcting outputs before approval
- How approvals flow into the self-improving loop
- Configuring review requirements based on risk level
- Review history and audit trail
Related
- Guardrails & Safety: how HITL fits into the broader safety architecture
- Review API: programmatic review endpoints
- Query Curation: managing approved queries via the Tune page