Skip to content

AIP-C01 · Study Guide

← Domain 3: AI Safety, Security, and Governance

3.1.3 · Lesson 3 of 15

Accuracy Verification & Hallucination Mitigation

Verify model response accuracy and prevent ungrounded hallucinations using grounding checks and structured output enforcement.

WHAT

Verify model response accuracy and prevent ungrounded hallucinations.

HOW

  • Contextual Grounding Checks (Bedrock Guardrails): Evaluate completions against source passage reference chunks for factual alignment.
  • Structured Output Enforcement: Use JSON Schema, Converse API tool parameters, or Lambda validators to force deterministic structures.

Contextual Grounding requirement

Requires passing both the candidate output and the retrieved source reference passages.

Retrieved Chunks (reference)
        +
Candidate Completion
        ↓
Contextual Grounding Check
        ↓
Grounded / Ungrounded decision

EXAM TRAP

Contextual Grounding does not evaluate candidate completions against the user prompt string alone.

Key points

  • Contextual Grounding Checks
  • Candidate output + source reference passages
  • Structured output / JSON Schema
  • Converse API tool parameters
  • Lambda validators for deterministic structure

Exam focus

Contextual Grounding requires both the generated candidate and retrieved source reference passages — it does not evaluate against the user prompt alone.

View progress