Task 1.6: Prompt Engineering & Governance

Task 1.6: Prompt engineering strategies và governance

Skill 1.6.1: Model instruction frameworks

Kiến thức cần nắm:

• Amazon Bedrock Prompt Management — Enforce role definitions, parameterized templates
• Amazon Bedrock Guardrails — Enforce responsible AI guidelines
• System prompts, user prompts, assistant prompts
• Template configurations cho response formatting

Skill 1.6.2: Interactive AI systems

Kiến thức cần nắm:

• Step Functions cho clarification workflows
• Amazon Comprehend cho intent recognition
• DynamoDB cho conversation history storage
• Multi-turn conversation management

Skill 1.6.3: Prompt management và governance

Kiến thức cần nắm:

• Amazon Bedrock Prompt Management — Parameterized templates, approval workflows
• Amazon S3 — Template repositories
• AWS CloudTrail — Track prompt usage
• Amazon CloudWatch Logs — Log prompt access

Skill 1.6.4: Quality assurance cho prompts

Kiến thức cần nắm:

• Lambda functions để verify expected output
• Step Functions để test edge cases
• CloudWatch để test prompt regression
• A/B testing cho prompt variants

Skill 1.6.5: Nâng cao FM performance qua prompt refinement

Kiến thức cần nắm:

• Structured input components
• Output format specifications (JSON mode, XML tags)
• Chain-of-thought instruction patterns
• Few-shot prompting
• Feedback loops cho iterative improvement

Skill 1.6.6: Complex prompt systems

Kiến thức cần nắm:

• Amazon Bedrock Prompt Flows — Sequential prompt chains
• Conditional branching dựa trên model responses
• Reusable prompt components
• Pre-processing và post-processing steps

Tài liệu tham khảo

• Amazon Bedrock Prompt Management
• Amazon Bedrock Prompt Flows
• AWS Blog - Prompt Engineering Best Practices
• Anthropic Prompt Engineering Guide