Chatbot example with Agentic RAG for an insurance company

This chatbot is built using an Agentic RAG approach and FastAPI. It leverages the Mixtral-8x7B model with Chroma database and incorporates three tools corresponding to different RAG implementations: FAQ queries, product details queries, and occupation-based insurance product queries.

* The Mixtral-8x7B model is free to run. You may improve the overal performance of the agent using a more powerful LLM such as GPT 3.5 or later.

Diagram

Sample run demo

Features

• Interactive chat interface
• Agentic RAG
• Web-based UI using FastAPI and Jinja2 templates
• Chroma DB

Prerequisites

• Python 3.11+
• FastAPI
• LangChain
• LangSmith API key
• NVIDIA AI API key
• Uvicorn

Project Structor

project_root/
│
├── assets/
│   ├── chat_screen.webm
│   └── diagram.jpg   
├── data/
│   ├── raw:JSON, CSV
│   └── vector: ChormaDB
├── src/
│   ├── __init__.py
│   ├── agent.py
│   ├── config.py
│   ├── data_classes.py
│   ├── db_creator.py
│   ├── rags.py
│   ├── utils.py
│   └── templates/
│       └── indexes.html
│── tests/
│   ├── __init__.py
│   └── test_rag.py      
├── README.md
├── Dockerfile
└── requirements.txt

Installation

To locally run the agent use the following steps

1. Clone the repository:

git clone https://github.com/YJonmo/agentic_rag.git
cd agentic_rag

2. Create a virtual env:

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

3. Get an API key from Nvidia website and an API key for LangSmith and place them in your .env file using the following commands:

echo NVIDIA_API_KEY="your-nvidia-api-key" >> .env
echo LANGSMITH_API_KEY="your-langsmith-key" >> .env

4. Run the following file to create a vector DB from the JSON and the CSV files in the data/raw folde:

python src/db_creator.py

5. Test the vector DB:

pytest

If there was not error then the vector DB is created inside the .data/vector folder.

6. Run the agent locally:

uvicorn src.agent:app --host 0.0.0.0 --port 8000 --reload

7. Open a web browser at:

http://0.0.0.0:8000/

You could build the docker image and deploy either locally or in the cloud

1. Build the docker image:

docker build -t agentic_rag .

2. Run the docker image:

docker run -e NVIDIA_API_KEY="your-nvidia-api-key" -p 8000:8000 agentic_rag