METEORA

Traditional RAG pipelines rely on similarity-based re-ranking with arbitrary top-k cutoffs: opaque, rigid, and vulnerable to adversarial content. METEORA replaces the re-ranking step entirely: a LLM generates rationales that guide evidence selection, explain every decision, and power a verifier that filters poison that filters poisoned chunks.

METEORA is a drop-in reranker replacement for retrieval-augmented generation pipelines. Instead of choosing a fixed top-k with a reranker, METEORA asks a language model to generate query-specific search rationales, then uses those rationales to select the evidence chunks that should be sent to the final generator.