Rag Application with Multi-Agent Orchestrator

This repository is an advanced implementation of the Retrieval-Augmented Generation (RAG) framework combined with multi-agent orchestration techniques. It integrates various agentic patterns such as Planning (ReAct flow), Reflection, and Multi-Agent workflows to enhance response generation and contextual understanding.

Table of Contents

1. Introduction to RAG
2. Multi Agent Orchestrator
3. Advanced RAG Techniques
4. Running Frontend or Backend Only
5. Running the Entire Project with Docker and Docker Compose
6. Project Structure
7. Contributing
8. License
9. References

---

1. Introduction to RAG

Large Language Models are trained on a fixed dataset, which limits their ability to handle private or recent information. They can sometimes "hallucinate," providing incorrect yet believable answers. Fine-tuning can help, but it is expensive and not ideal for frequent updates. The Retrieval-Augmented Generation (RAG) framework addresses this issue by using external documents to improve the LLM's responses through in-context learning. RAG ensures that the information provided by the LLM is not only contextually relevant but also accurate and up-to-date.

There are four main components in RAG:

1. Indexing: Documents are split into chunks, and embeddings for these chunks are created and stored in a vector database.
2. Retriever: The retriever finds the most relevant documents based on the user's query using vector similarity search.
3. Augment: The retrieved documents are combined with the user query to form a prompt that provides contextual information.
4. Generate: The prompt is fed into the LLM to generate an accurate and context-aware response.

---

2. Multi Agent Orchestrator

This repository implements multi-agent workflows that enhance LLM capabilities through agent collaboration. It integrates:

• ReAct Flow for planning and execution
• Reflection Mechanisms to improve agent performance
• Multi-Agent Coordination for complex problem-solving

How It Works

1. User input is classified to determine the appropriate agent.
2. The orchestrator selects the best agent based on historical context and agent capabilities.
3. The selected agent processes the input and generates a response.
4. The orchestrator updates conversation history and returns the response to the user.

For further exploration:

• Agentic Patterns Repo
• Multi-Agent Orchestrator

---

3. Advanced RAG Techniques

This repository supports several advanced RAG techniques:

Technique	Tools	Description
Naive RAG	LlamaIndex, Qdrant, Google Gemini	Basic retrieval-based response generation.
Hybrid RAG	LlamaIndex, Qdrant, Google Gemini	Combines vector search with traditional methods like BM25.
Hyde RAG	LlamaIndex, Qdrant, Google Gemini	Uses hypothetical document embeddings to improve retrieval accuracy.
RAG Fusion	LlamaIndex, LangSmith, Qdrant, Google Gemini	Generates sub-queries, ranks results with Reciprocal Rank Fusion, and improves retrieval performance.
Contextual RAG	LlamaIndex, Qdrant, Google Gemini, Anthropic	Compresses retrieved documents to keep only the most relevant details.
Unstructured RAG	LlamaIndex, Qdrant, FAISS, Google Gemini, Unstructured	Handles text, tables, and images for more diverse content retrieval.

---

4. Running Frontend or Backend Only

• To run the backend separately, follow the instructions in the backend README.
• To run the frontend separately, follow the instructions in the frontend README.

---

5. Running the Entire Project with Docker and Docker Compose

5.1 Prerequisites

• Install Docker
• Install Docker Compose

5.2 Steps

1. Clone the Project

git clone https://github.com/buithanhdam/rag-app-agent-llm.git
cd rag-app-agent-llm

2. Configure Environment Variables

cp ./frontend/.env.example ./frontend/.env
cp ./backend/.env.example ./backend/.env

and fill:

# For backend .env
GOOGLE_API_KEY=<your_google_api_key>
OPENAI_API_KEY=<your_openai_api_key>
ANTHROPIC_API_KEY=<your_anthropic_api_key>
BACKEND_API_URL=http://localhost:8000
QDRANT_URL=http://localhost:6333

MYSQL_USER=your_mysql_user
MYSQL_PASSWORD=your_mysql_password
MYSQL_HOST=your_mysql_host
MYSQL_PORT=your_mysql_port
MYSQL_DB=your_mysql_db
MYSQL_ROOT_PASSWORD=root_password

AWS_ACCESS_KEY_ID=
AWS_SECRET_ACCESS_KEY=
AWS_REGION_NAME=
AWS_STORAGE_TYPE=
AWS_ENDPOINT_URL=

# For frontend .env
NEXT_PUBLIC_BACKEND_API_URL=http://localhost:8000

3. Build and Run the Project

docker-compose up --build

4. Set Up MySQL Database

docker exec -it your-container-name bash
mysql -u root -p

• Enter root password (configured in .env or docker-compose.yml).

Run SQL queries:

CREATE USER 'user'@'%' IDENTIFIED BY '1';
GRANT ALL PRIVILEGES ON ragagent.* TO 'user'@'%';
FLUSH PRIVILEGES;
CREATE DATABASE ragagent;

5. Access the Application

• Frontend: http://localhost:3000
• Backend: http://localhost:8000
• Qdrant: Exposes ports 6333, 6334
• MySQL: Exposes port 3306

6. Stop the Project

docker-compose down

---

6. Project Structure

• backend/: Backend source code

  • Dockerfile.backend: Backend container setup
  • requirements.txt: Backend dependencies

• frontend/: Frontend source code

  • Dockerfile.frontend: Frontend container setup
  • next.config.js: Next.js configuration

• docker-compose.yml: Docker Compose setup

• Jenkinsfile: CI/CD configuration

---

7. Contributing

Contributions are welcome! Please submit an issue or a pull request to improve this project.

---

8. License

This project is licensed under the MIT License.

---

9. References

• Agentic Patterns Repo
• Multi-Agent Orchestrator
• kotaemon
• multi-agent
• RAG Cookbook