Controllable RAG Agent
Basic Information
This repository provides an advanced Retrieval-Augmented Generation (RAG) agent implementation designed to solve complex, non-trivial questions from user documents that simple semantic retrieval cannot handle. It demonstrates a deterministic graph that functions as the agent's "brain" and a controllable autonomous pipeline for processing PDFs, generating chapter summaries, creating a quotes database, encoding content into vector stores, and orchestrating multi-step question answering. The project includes a step-by-step notebook, a real-time Streamlit visualization, Docker and non-Docker run instructions, and integration examples using LangChain, FAISS, and various LLM providers. It is intended as a research and demonstration repo for controlled RAG workflows, showing how to anonymize questions, plan and decompose tasks, verify grounding in source data, and evaluate outputs using Ragas metrics.
Links
Categorization
App Details
-
$$$$ -
$$$$