mini-rag-app

Minimal RAG Application. This is my implementation of the mini-rag-app playlist by Eng. Abu Bakr Soliman on Youtube. Original Github repository: mini-rag.

Retrieval-Augmented Generation (RAG)

The aim is to provide additional materials for the user query, which will assist the AI model in generating appropriate responses.

Setup

This environment is setup to work on a Linux platform. Make sure to use WSL2 on windows.

• Clone this repository.

git clone https://github.com/e-hossam96/mini-rag-app.git

• Install developer tools for C++ package building.

sudo apt update
sudo apt upgrade -y
sudo apt install build-essential

• Download and install Miniconda

wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh -b -p ./miniconda

• Activate conda base environment.

source ./miniconda/bin/activate

• Create mini-rag env from YAML file

cd mini-rag-app/src
conda env create -f environment.yml
conda activate mini-rag

• Fill .env.example file and save it into a .env file. Notice, it is already filled with the default settings except the API keys. Feel free to change them to your preferences.

cp .env.example .env

• Configure Docker setups to run MongoDB container and deploy the application later.

  • Install docker. Details for Linux can be found here.
  • Add your user to docker group by following the steps in here.
  • Fill .env.example file using desired mongodb credentials and save it into a .env file.

  cp ../docker/.env.example ../docker/.env

  • Compose the docker services.

  docker compose up -d

• Run the FastAPI server. Use the --reload argument only for development. A Postman collection is available in the assets directory for your help.

uvicorn main:app --reload --host 0.0.0.0 --port 8000

Usage

You can use the exported Postman collection provided here and the sample text files to send queries to the different endpoints and evaluate the results.

Front End

The current front end code contain buttons to send requests to 4 routes out of the six defined so far.

To start the front end server, navigate to to src/views/ and run the following command.

python -m http.server 5000

This will launch the server on port 5000 which can be accessed using the URL http://localhost:5000

Some interactions has been recorded as follows:

• Sample Document (PDF research paper)

• Application Interface

• MongoDB with Text Chunks