This repository serves as an example for creating a streamlit application. The goal is to be able to quickly spin up a RAG chatbot (with a streamlit interface) that can answer questions about your own data.

To use the code in this repo, there are two things you need: data (duh) and a OpenAI API key.

First, provide the key in a file you call config.ini in this format:

[credentials]
api_key = <your_OpenAI_API_key>

Then, you must prepare your data. Using Langchain, a locally stored vector database will be created from the data. See src/create_vdb.py for implementation. So in essence:

1. Create a csv file where each row are text chunks, and contains the text in a column named "text".
2. Name the file <your_data>.csv and save it in the data folder.
3. Set KNOWLEDGE_BASE = "your_data" in create_vdb.py and in app.py

Currently, there is an example for "hjertemedisin" (heart medicine) data, and you can see how the data is stored as a csv in data/hjertemedisin.csv

Before starting, make sure you have Python 3.x installed on your system.

Clone this repository to your local machine using Git:

git clone https://github.com/martinstiles/RAG-Streamlit-Chatbot
cd RAG-Streamlit-Chatbot

Initialize the python environment by running the following command inside the project folder.

python -m venv venv

Activate the venv by running the following on Windows:

venv\Scripts\activate

or on Mac / Linux:

source venv/bin/activate

All required python packages are found in requirements.txt and can be installed by running:

pip install -r requirements.txt

If you have followed the setup procedure, create a vector database by running src/create_vdb.py.

If you have created a vector database, the project can be run with the following command:

streamlit run src/app.py

If you want, you can run the applicaiton in a Docker container. First, make sure you have Docker installed by running:

docker --version

To create a Docker image for the streamlit application on your local machine and run the following command:

docker build -t llm_demo .

Once the image has been created, you can run a container locally by running:

docker run -p 8501:8501 llm_demo

In general, you should never feed sensitive data into a public API. If you are trying to chat with your own private data and use a LLM API to accomplish it (like in this repo): make sure the data are non-sensitive.