Agentic-AutoRAG

Reasoning-driven optimizer for RAG pipelines: LLM agents diagnose retrieval failures and propose the next config.

Agentic AutoRAG is a reasoning-driven optimizer for Retrieval-Augmented Generation (RAG) pipelines. Instead of grid search or Bayesian optimization, it runs a two-stage LLM agent loop: a diagnoser analyses why a trial configuration fails, and a proposer chooses what to change next from that diagnosis and the history of prior trials. The optimization signal is a synthetic exam, open-ended questions with ground-truth answers generated from your corpus on the first run and cached for reuse.