What is the difference between basic RAG and Agentic RAG?

Basic RAG follows a linear retrieve-and-generate path, while Agentic RAG uses decision-making loops to evaluate document quality and retry or expand searches if needed.

What is Corrective-RAG (CRAG)?

CRAG is a pattern that grades retrieved documents for relevance and triggers a web search fallback if the local knowledge base doesn't provide sufficient information.

Does this skill require LangGraph?

Yes, it is optimized for LangGraph 1.0.6+, utilizing its state management and conditional routing features for robust, production-ready workflows.

How does query transformation help in retrieval?

It rephrases user queries based on feedback from the retrieval stage to improve the chances of finding relevant documents in subsequent attempts.

Agentic RAG Patterns

Name: Agentic RAG Patterns
Author: yonatangross

byyonatangross

•

Data Science & ML

Builds self-correcting RAG systems using LangGraph for adaptive retrieval, document grading, and web search fallbacks.

Agentic RAG Patterns provides a comprehensive set of architectural blueprints for creating sophisticated retrieval systems that go beyond basic vector search. By leveraging LangGraph 1.0.6, this skill enables developers to implement Self-RAG and Corrective-RAG (CRAG) workflows, where the AI autonomously grades document relevance, rewrites poor queries, and falls back to web searches when necessary. It is ideal for building production-ready AI agents that require high-precision data retrieval and the ability to self-correct during the reasoning process to minimize hallucinations and improve response accuracy.

Key Features

01CRAG document grading and automated web search fallback

02Self-RAG adaptive retrieval logic with reflection tokens

03Knowledge Graph and Vector hybrid retrieval (GraphRAG) support

04Query transformation and multi-step rephrasing loops

05LangGraph-driven state management and conditional routing

0669 GitHub stars

Use Cases

01Building agentic workflows that require real-time web research when local data is insufficient

02Implementing multi-hop reasoning pipelines using combined Knowledge Graphs and vector stores

03Developing high-precision Q&A bots for complex technical documentation or internal wikis

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add yonatangross/orchestkit agentic-rag-patterns

For use in Claude.ai and ChatGPT

Download Skill