Keyword search is all you need: Achieving RAG-Level Performance without vector databases using agentic tool use

This feed contains multiple 2026 CS/IR papers focused on retrieval, RAG, and recommendation systems. Key contributions include: an agentic keyword-search approach that attains >90% of RAG performance without persistent vector DBs; TTE-v2, a reasoning-driven multimodal retrieval and reranking framework; domain-partitioned hybrid RAG with a Neo4j legal knowledge graph for Indian legal QA; SPRIG, a CPU-only GraphRAG using NER co-occurrence graphs + PPR; Higress-RAG, an enterprise "full-link" RAG stack with adaptive routing, semantic caching and corrective RAG (CRAG); Geodesic Semantic Search that