Beyond Chunk and Pray: Smarter RAG for Drupal AI
25 Mar 2026, 9:10 pm (UTC)
50 Minutes
Speakers
For a Drupal site with a hundred thousand nodes, the standard approach of splitting content into chunks and storing them in a vector database calls for a different retrieval-augmented generation (RAG) strategy. The session introduces RAG and embeddings, which combine large language models with external knowledge.
Topics include hybrid retrieval that combines semantic embeddings with BM25 keyword search, chunking based on headings, paragraphs and semantic cues, and re-ranking with smaller models. The session also covers using user profiles, session data and relationships between nodes to provide context to the model.