Query transformation: rewrite follow-ups into standalone queries, generate multiple query variants, or HyDE (embed a hypothetical answer to retrieve similar passages).
Better chunking: structure-aware splits, parent-child (retrieve small chunks, return larger context), and metadata enrichment.
Corrective and agentic RAG: grade retrieved results, re-query or search the web when they are weak, and let an agent decide when and what to retrieve.
Graph RAG for relationship-heavy data, and compressing or reordering retrieved context before generation.
Loading…