AgenticRAG: Deep Research Framework with Agentic RAG

AgenticRAG implements a powerful Deep Research Framework that combines intelligent query planning with reliable Retrieval-Augmented Generation (RAG) for comprehensive question answering. The system takes complex queries and automatically breaks them down into multiple focused sub-queries, processes each through the full Agentic RAG pipeline, and synthesizes all findings into comprehensive reports.

Deep Research Framework (Main System)

The core innovation is a deep research framework that orchestrates comprehensive analysis:

Deep Research Framework with AgenticRAG
│
├── Search Planner (generate multiple focused queries)
│
├── For each query:
│   ├── AgenticRAG Processing
│   │   ├── Query Routing
│   │   ├── RAG Pipeline (with all reliability features)
│   │   │   ├── Document Retrieval
│   │   │   ├── Document Grading
│   │   │   ├── Query Rewriting (if needed)
│   │   │   └── Web Search Fallback
│   │   └── Answer Generation
│   └── Store individual answer
│
└── Final Report Generator (aggregate all answers)

How It Works

1. Intelligent Query Planning: Automatically generates 3-5 diverse, focused search queries from complex topics
2. Reliable Processing: Each sub-query is processed through the full Agentic RAG pipeline with all reliability features
3. Comprehensive Synthesis: Results are aggregated into a well-structured, comprehensive report

Example Workflow

For a query like "What are the latest advancements in quantum computing?":

1. Query Planning: Generates focused sub-queries:

   • "Recent breakthroughs in quantum hardware and qubit stability"
   • "Latest quantum algorithms for optimization and machine learning"
   • "Current industry applications of quantum computing"

2. Agentic RAG Processing: Each sub-query is processed with:

   • Intelligent routing (direct response, RAG, or web search)
   • Document retrieval and relevance grading
   • Query rewriting when needed (up to 3 retries)
   • Web search fallback for time-sensitive information

3. Report Generation: Synthesizes all findings into a comprehensive report

System Architecture

The Agentic RAG system follows a modular architecture with the following components:

Prerequisites

• Python 3.8+
• Jupyter Notebook or JupyterLab
• Ollama with Qwen model (qwen3:8b)
• Serper API key for web search functionality

Installation

1. Clone the repository:

   git clone <repository-url>
   cd AgenticRAG

2. Install required dependencies:

   pip install langchain langchain-community langchain-huggingface langgraph faiss-cpu requests

3. Install Ollama and pull the Qwen model:

   # Install Ollama from https://ollama.com/
   ollama pull qwen3:8b

4. Set up your Serper API key:

   export SERPER_API_KEY=your_serper_api_key_here

Usage

Main Deep Research System

Open and run notebooks/DeepResearch_with_AgenticRAG.ipynb:

1. The system takes complex queries and automatically breaks them down into multiple focused sub-queries
2. Each sub-query is processed through the full Agentic RAG pipeline with all reliability features
3. Results are synthesized into a comprehensive report

Example complex query: "What are the latest advancements in quantum computing?"

The system will:

1. Generate focused sub-queries on different aspects
2. Process each through the reliable Agentic RAG system
3. Create a comprehensive report synthesizing all findings

Core Agentic RAG Features (Component)

The Agentic RAG system that powers each query processing includes:

Multi-Approach RAG Implementation

1. Adaptive RAG: Intelligently routes queries to the most appropriate processing path
2. Corrective RAG: Uses web search as a fallback when internal knowledge is insufficient
3. Self-RAG: Implements self-correction mechanisms to reduce hallucinations

Intelligent Query Routing

Routes queries to one of three processing paths:

• Direct Response: For simple greetings or general knowledge questions
• Vectorstore RAG: For domain-specific questions with document relevance grading
• Web Search: As a fallback when internal RAG fails

Advanced RAG Workflow

• Document relevance grading to ensure quality responses
• Query rewriting mechanism for improved retrieval on failed attempts
• Automatic fallback to web search after maximum RAG retries
• Stateful agent behavior with conversation history tracking

Customization

Knowledge Base Management

The notebooks include code to rebuild the FAISS vector database from source documents. You can customize the knowledge base by:

1. Modifying the source URLs in the database creation section
2. Adjusting chunk size and overlap parameters
3. Changing the embedding model

Model Configuration

You can customize the LLM by modifying the ChatOpenAI configuration:

• Change the model name
• Adjust temperature for creativity vs. consistency
• Modify other model parameters

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• This implementation is based on concepts from the LangChain documentation and tutorials
• The knowledge base is built from Lilian Weng's blog posts on LLM topics
• Uses FAISS for efficient vector similarity search
• Integrates with Serper API for web search capabilities