This repository provides a minimal, production-ready Agentic Retrieval-Augmented Generation (RAG) system built with LangGraph. It targets engineers and power users seeking to learn and implement advanced RAG capabilities like hierarchical indexing, conversation memory, and human-in-the-loop query clarification. The project bridges the gap between basic RAG tutorials and deployable applications, offering a modular and customizable framework.

How It Works

The system employs a four-stage intelligent workflow orchestrated by LangGraph. It utilizes hierarchical indexing, splitting documents into small "Child" chunks for precise retrieval and larger "Parent" chunks for contextual depth. Conversation memory maintains dialogue continuity, while an automated query clarification stage resolves ambiguity or prompts for human input. An agent orchestrates these components, performing self-correction and re-querying if initial results are insufficient, ensuring comprehensive and accurate responses.

Quick Start & Requirements

• Installation: Two paths are offered: an interactive notebook (Colab or local Jupyter/VSCode) or a full Python project. Both require installing dependencies via pip install -r requirements.txt. Users must place PDF files in the docs/ directory.
• Prerequisites: Python environment. Support for multiple LLM providers: Ollama (local, recommended for development), Google Gemini, OpenAI, and Anthropic Claude (requiring API keys for cloud services).
• Links: Documentation for PDF conversion techniques is available via a companion notebook. The main application can be run locally at http://127.0.0.1:7860.

Highlighted Details

• Dual Learning Paths: Offers both a step-by-step interactive notebook and a modular project structure for flexible learning and development.
• Provider-Agnostic LLMs: Seamlessly switch between Ollama, Gemini, OpenAI, and Claude with minimal code changes.
• Hierarchical Indexing: Combines the precision of small child chunks with the context of larger parent chunks for improved retrieval accuracy.
• Advanced Agentic Features: Includes conversation memory, intelligent query clarification (human-in-the-loop), and self-correction capabilities.
• Modular Architecture: Core components (LLM provider, agent workflow, document processing, embedding models) are independently swappable for customization.
• End-to-End Gradio Interface: Provides a complete interactive RAG pipeline with document management.

Maintenance & Community

Contributions are welcomed via issues or pull requests. An "Upcoming Features" section indicates ongoing development, with "Multi-Agent Map-Reduce" listed as "In Progress" for a December 2025 release.

Licensing & Compatibility

The project is released under the MIT License, permitting free use for learning and building personal projects.

Limitations & Caveats

The "Multi-Agent Map-Reduce" feature is currently in development. Specific unsupported platforms or known bugs are not detailed in the provided text.