// the find
asinghcsu/AgenticRAG-Survey
Agentic-RAG explores advanced Retrieval-Augmented Generation systems enhanced with AI LLM agents.
A companion repo to an arXiv survey paper on Agentic RAG — no code, just a long README cataloguing agent patterns (reflection, planning, tool use, multi-agent) and taxonomy for RAG architectures, plus a table linking out to third-party implementation notebooks. Useful for someone who wants a map of the RAG-agent landscape rather than something to install or run.
The implementation table is the actual payoff — it links each taxonomy category (single-agent, router-based, graph-based, etc.) to a specific working notebook from LangChain, LlamaIndex, NVIDIA, IBM, and others, so you can go from concept to running code in one click. The comparative table (Traditional RAG vs Agentic RAG vs ADW) is a decent quick reference when you're deciding which architecture fits a problem. It's grounded in a real paper (arXiv:2501.09136) rather than being an unsourced blog roundup.
There's no code in the repo itself — it's a markdown document with image assets, so 'stars: 1736' reflects interest in the survey, not usage of anything. Large chunks are repetitive: the abstract, intro, and taxonomy sections restate the same four patterns three times in slightly different words. It depends entirely on external notebooks staying up; if NVIDIA or IBM reorganizes their cookbook repos, half the value here rots silently with no way to know from this repo. Ironically for a project warning against 'AI slop,' the prose leans hard on the exact hedge-everything, bullet-heavy style (bold key terms, emoji headers, 'comprehensive,' 'insights') that makes surveys like this hard to skim for a real decision.