You are implementing multi-agent RAG for a legal discovery suite. A Searcher agent retrieves passages; an Analyst agent reasons over them. Early tests show poor retrieval precision and lost clause boundaries. Which RAG design change best addresses embedding quality and retrieval precision?
Select an answer to reveal the explanation.
Short Explanation
Legal retrieval lives or dies on chunk boundaries and retrieval precision. Keep clauses intact with structure-aware chunking, use solid domain-appropriate embeddings, and tune hybrid/semantic retrieval with matter/privilege metadata. Tiny shards shred meaning, titles-only under-retrieves, and stuffing whole folders isn't multi-agent retrieval.
Full Explanation
Multi-agent retrieval-augmented generation architecture skills include chunking, embedding quality, and retrieval precision. The structure-aware chunking approach with high-quality domain-appropriate embeddings and tuned retrieval—using filters, hybrid/semantic ranking, and matter/privilege metadata—improves each dimension and preserves privileged-matter controls via metadata filters. The fixed 20-token fragment approach destroys semantic units critical in legal text. The titles-only embedding approach yields high precision only on titles while missing operative clause content. The whole-folder stuffing approach abandons retrieval architecture, blows context limits, and increases cost and latency. Exam tip: Specialist retrieval agents still need a high-quality index design underneath the orchestration.