Controllable-RAG-Agent

Why we included this project

Plain semantic-search RAG answers straightforward questions well, but anything multi-hop or conditional tends to come back wrong. This repository replaces the naive similarity lookup with a hand-crafted deterministic graph that acts as the agent's reasoning layer, deciding which sources to consult, when to branch, and how to chain sub-queries before writing an answer. The code ships as self-contained Jupyter notebooks on LangChain and LangGraph, so you can run the examples, trace the flow, and reuse the pattern in your own pipeline instead of treating it as a black box. It is a solid reference implementation for developers who want controllable, step-by-step RAG on their own data, and the companion book goes deeper into why each technique works.