AI Playbooks

Playbooks turn the atlas into operating practice. Use them when you need a repeatable procedure with inputs, steps, outputs, and a definition of done.

Mental model

A playbook is an operational interface between evidence and action. It names prerequisites, inputs, ordered checks, decision gates, artifacts, rollback conditions, and a verifiable definition of done so the result does not depend on who remembers the process.

Roadmap: evaluation and quality

• Evaluate RAG answer quality
• Build an eval set from scratch
• Debug a hallucination
• Ship a prompt change safely

Architecture and delivery

• Decide prompt vs RAG vs fine-tune
• Choose a model for production
• Measure and cut inference cost
• Stand up LLM observability

Agents and security

• Debug an agent stuck in a loop
• Add a human approval gate
• Audit an app for prompt injection
• Run an AI red team lite

Connects to: Evaluation · MLOps · AI Safety and Security

Core sources

• NIST AI RMF Playbook — suggested actions that operationalize risk-management outcomes.
• OWASP GenAI Security Project — testable application and agent security risks.
• RAGAS documentation — concrete RAG evaluation metrics and workflows.
• Google Rules of ML — concise production checks and sequencing principles.