Human Review and Editing

After automated checks, content may be reviewed and edited by a human or through an agent-assisted editing workflow.

Human-in-the-loop (HITL)

HITL is the process where a human reviewer evaluates generated content and decides whether it meets quality standards. The reviewer can:

Approve the content as-is
Request changes with specific feedback
Reject and trigger regeneration from scratch

HITL applies to all content modalities: text, images, video, questions, and artifacts.

Regenerate vs edit

When content needs improvement, there are two approaches:

Regeneration

Generate the content again from the same blueprint, potentially with adjusted parameters or additional context. Best when:

The content fundamentally misses the mark
The structure or approach needs to change
The issue is in the concept, not the execution

Editing

Modify the existing content while preserving its structure and intent. Best when:

Specific factual corrections are needed
Wording or tone adjustments are required
Minor structural changes (reorder, trim, expand a section)

Agent-supported editing

Fraya operates in two modes:

Autopilot — full course generation through the pipeline, with automated checks
Co-pilot — human-guided editing where the agent assists with specific changes

In co-pilot mode, the agent can:

Analyze existing content and suggest improvements
Apply targeted edits based on human feedback
Regenerate individual sections while preserving course consistency
Perform bulk operations across multiple sections

For the technical design of the editing system (multi-model architecture, tool inventory, database schema, orchestration flows), see docs/fraya/course-editing-design.md.