Build grounded retrieval with controlled ingestion, citations, access enforcement, evaluation, lineage, and safe no-answer behavior.
Learning outcomes
Curriculum
The chapters are individually addressable documentation pages. You can link directly to a concept from another program, architecture decision, or implementation guide.
Define the RAG system, its boundaries, and the knowledge lifecycle.
Design retrieval as a governed knowledge system rather than a prompt attachment.
Set the intended purpose, actors, assets, dependencies, lifecycle, and system boundary before selecting controls.
Map where trust changes and where data, artifacts, requests, identities, tools, and evidence cross a boundary.
Maintain lineage and enforce the source authorization model throughout retrieval.
Preserve source identity, classification, ownership, permission, version, and transformation lineage.
Enforce source permissions through ingestion, indexing, retrieval, caching, generation, and evidence handling.
Preserve user, workload, service, and tool identity across AI boundaries and make policy decisions explicit.
Design retrieval, reranking, citation, grounding, and no-answer contracts.
Design retrieval and answer contracts that expose support, uncertainty, and safe refusal behavior.
Test relevance, support, permissions, injection resistance, stale content, and adversarial cases.
Turn acceptance criteria into repeatable promotion decisions with recorded evidence and rollback conditions.
Observe behavior, control change, retain evidence, and record architecture choices.
Design signals that explain service health, AI behavior, policy outcomes, change, and incidents without leaking sensitive content.
Version complete AI behavior, assess change impact, approve promotion, support rollback, and retire superseded assets.
Design evidence alongside controls so important decisions, releases, configurations, and operating events can be verified.
Record context, options, control consequences, evidence, dependencies, limitations, and reversal triggers for material decisions.
Build the evaluation package and verify the module outcomes.
Run a synthetic local retrieval system that enforces document permissions, citations, deletion, no-answer behavior, and injection controls.
Runnable synthetic prototype with recorded validation
Public exercises use synthetic or explicitly approved documents. The curriculum does not authorize indexing customer data or claim that one retrieval stack is suitable for every workload; tested implementation details remain adapter-specific.