← Back
Process Design · AI-Assisted QA

A decade of content and no way to review it all at human pace.

I designed a four-stage workflow where AI handles extraction and flagging, and the subject matter expert focuses on the judgment calls. The system helps support the human reviewer.

01

Human review is essential, but it doesn't scale.

A large content catalog spanning two platforms. The quality audit found that many products, some over a decade old, didn't meet current requirements. Expert review could identify the issues for remediation, but there wasn't enough expert time to review everything.

Reviewed at human pace
Waiting
02
AI Briefings,
Human Decisions.

The AI never replaces the expert. It produces a briefing that helps the expert review only the problems.

03

Four-stage review where the judgment stays human.

The model's review is guided by detailed process instructions and supported by content frameworks and guidelines.

1
Structural Inventory
AI-driven
+
  • Map content to requirements at page level
  • Compile key terms with source locations
  • Build citations inventory across all material
2
Analytical Flagging
AI-driven, expert-designed
+
  • Coverage confidence: depth versus presence
  • Gap and bridge opportunities for new material
  • Quality flags: outdated content, sensitivity concerns
  • Accessibility: language complexity, vocabulary load
3
Expert Deep Read
Human-driven, AI-informed
+
  • Evaluate whether AI assessments hold up
  • Triage flagged issues: real problems or false positives
  • Catch what the AI missed
4
Synthesis & Documentation
Human-driven
+
  • Final coverage map: met, partial, or missing
  • Prioritized remediation recommendations
  • Documentation that accelerates the next cycle
04

Drawing the line.

The workflow redraws where human attention goes.

Shifts to AI
Locating where requirements appear across hundreds of pages
Stays with the expert
The deep read and all coverage judgments
Shifts to AI
Compiling documentation alongside development
Stays with the expert
Designing the evaluation framework and quality criteria
Shifts to AI
Initial scan for coverage and quality flags
Stays with the expert
Sensitivity and bias assessment

Each expert tailors the evaluation framework to their content area.

05

Directing expert time and attention.

An expert without a briefing has to read everything to get a sense of the gaps. With the AI providing an initial review, the expert knows where to review.

Every page looks the same. The expert reads all of it with equal attention, hunting for where problems might be.

The briefing marks what needs scrutiny. The expert's time goes to the key judgment calls.

Gap identified
Quality flag
Coverage confirmed
06

Scaling development and quality at the same time requires directing expert attention where it is actually needed.

Designed and proposed as a pilot process. Not a shipped implementation with measured results.