Agentic Retrieval-Augmented Generation as a Knowledge Representation Layer

Agentic retrieval-augmented generation resolves retrieval policy, context construction and safety adjudication dynamically, per conversational turn. Grounding, provenance and reproducibility are therefore determined by the composition of the pipeline itself. This lecture characterises corpus segmentation, indexing, retrieval and context assembly as a single knowledge representation layer, which delimits the propositions an agent can ground prior to generation, and treats its parameterisation as a design space open to empirical study.