Advanced Agentic RAG System

A powerful, multi-agent Retrieval-Augmented Generation (RAG) application built with Streamlit, LangChain, and Groq LLMs. This project demonstrates a modular, self-correcting workflow for answering user queries using both internal knowledge sources and real-time web search, with built-in fact-checking and safety mechanisms.

---

Key Features

• Multi-Agent Workflow: Modular agents for routing, retrieval, query reformulation, web search, synthesis, answer generation, fact-checking, and safety checking.
• Hybrid Knowledge Sources: Supports both uploaded files (PDF, DOCX, TXT) and URLs as knowledge bases, with fallback to default sources.
• Dynamic Query Handling: Automatically routes queries to the best agent (internal retrieval, web search, clarification, etc.) based on context.
• Self-Correction: Reformulates queries and retries retrieval before falling back to web search.
• Fact-Checking: Extracts factual claims from generated answers and verifies them using real-time web search.
• Safety Checking: Analyzes generated content for harmful or inappropriate material and revises or blocks unsafe responses.
• Interactive UI: Streamlit-based chat interface with workflow visualization, logs, and configuration sidebar.
• Configurable Parameters: Easily adjust chunk size, retriever top-K, and LLM temperature from the sidebar.

---

Architecture and Workflow

flowchart TD
    Start([Start]) --> Router
    Router -->|retrieve| Retriever
    Router -->|reformulate| Reformulator
    Router -->|web_search| WebSearcher
    Router -->|clarify| Clarifier
    Router -->|generate| Generator
    Retriever --> Grader
    Grader -->|relevant| Synthesizer
    Grader -->|reformulate| Reformulator
    Grader -->|web_search| WebSearcher
    Grader -->|clarify| Clarifier
    Reformulator --> Retriever
    Reformulator -->|fallback| WebSearcher
    WebSearcher --> Synthesizer
    Synthesizer --> Generator
    Generator --> FactChecker
    FactChecker --> SafetyChecker
    SafetyChecker --> End([End])
    Clarifier --> End

Loading

Agent Roles:

• Router: Decides the best workflow path for each query.
• Retriever: Fetches relevant documents from the internal knowledge base.
• Reformulator: Improves queries for better retrieval.
• WebSearcher: Finds real-time, external information.
• Synthesizer: Combines information from various sources.
• Generator: Creates the final answer.
• FactChecker: Verifies factual claims in the generated answer.
• SafetyChecker: Ensures generated content is safe and appropriate.
• Clarifier: Asks user for more details if needed.

---

Installation and Setup

1. Clone the Repository

git clone https://github.com/sainikhil-m/advanced-agentic-rag.git
cd advanced-agentic-rag

2. Create a Virtual Environment

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install Dependencies

pip install -r requirements.txt

Note: This project requires Python 3.9+.

4. Set Up API Keys

Create a .streamlit/secrets.toml file in the project root with the following content:

LANGCHAIN_API_KEY = "your_langchain_api_key"
TAVILY_API_KEY = "your_tavily_api_key"
GOOGLE_API_KEY = "your_google_api_key"
GROQ_API_KEY = "your_groq_api_key"

• Get your API keys from:
  • LangChain (https://smith.langchain.com/)
  • Tavily (https://app.tavily.com/)
  • Google Generative AI (https://ai.google.dev/)
  • Groq (https://console.groq.com/)

5. Run the Application

streamlit run app.py

---

Usage

1. Configure Knowledge Sources:

   • Add URLs (one per line) and/or upload files (TXT, PDF, DOCX) in the sidebar.
   • Adjust chunk size, retriever K, and LLM temperature as needed.
   • Click "Apply Parameters & Update Knowledge" to refresh the knowledge base.

2. Chat:

   • Enter your question in the chat input.
   • The system will process your query through the multi-agent workflow and display the answer.
   • View workflow logs, knowledge sources, and a diagram of the agent flow in the "Execution Details" section.

3. Reset:

   • Use "Clear Chat History" to reset the conversation and logs.

---

Requirements

• Python 3.9+
• Streamlit
• LangChain
• Groq LLMs
• Tavily API
• Google Generative AI API
• ChromaDB
• Other dependencies as listed in requirements.txt

---

Customization

• Add More Agents: Extend the workflow by adding new agent nodes and decision logic.
• Change LLMs: Swap out LLM providers or models by modifying the agent functions.
• Integrate More Tools: Add new retrieval or search tools as needed.

---

License

MIT License

---

Acknowledgements

• LangChain (https://github.com/langchain-ai/langchain)
• Streamlit (https://streamlit.io/)
• Groq (https://groq.com/)
• Tavily (https://tavily.com/)
• Google Generative AI (https://ai.google.dev/)

---

About the Developer

Sai Nikhil Mattapalli AI/ML Engineer

Sai is an AI/ML Engineer with 5 years of experience building and deploying scalable Machine Learning and Generative AI solutions across healthcare and IT domains. He specializes in end-to-end ML pipelines, LLMs, RAG, LangChain, and vector databases. This project serves as a demonstration of advanced multi-agent orchestration for production-grade AI applications.

• Email: ms1104n@gmail.com
• GitHub: https://github.com/sainikhil-m
• LinkedIn: https://www.linkedin.com/in/sainikhil-m

Enjoy building with advanced agentic RAG!