AI Act Law Retrieval Augmented Generation with LangChain

This project provides a conversational interface for users to ask questions about the AI Act Law using a collection of PDFs as reference. It leverages the following technologies:

• Backend (FastAPI):
  • Processes uploaded PDFs and extracts text.
  • Creates a vector store for efficient retrieval of relevant information from the extracted text.
  • Uses LangChain to handle the conversation flow and generate responses to user queries.
  • LangChain integrates retrieval augmentation, where retrieved information from the PDFs is used to contextually tailor the responses from the large language model (LLM).
• Frontend (Streamlit):
  • Provides a user-friendly chat interface for users to ask questions.
  • Uploads and processes PDFs through the backend API.
  • Displays the conversation history and the LLM's generated responses.

Key Features:

• Expertise in AI Act Law: The LLM was given information related to the AI Act Law, making it a valuable resource for users seeking expert advice on the topic. You can modify the prompt.py file + the PDFs you upload to serve a different task
• Retrieval Augmentation: By incorporating retrieved information from the uploaded PDFs, the LLM can generate more focused and relevant responses to user queries.
• Conversational Interface: The chat-based interface allows for a natural and interactive way to explore the information contained within the PDFs.

Dockerized Deployment:

The project utilizes Docker Compose to manage and deploy the backend and frontend services as a single application. This allows for easy deployment and ensures all required components run seamlessly together.

CI/CD Pipeline

This project leverages GitHub Actions to automate code linting on every push to the repository. The .github/workflows/main.yml file defines a workflow that runs the following jobs.

1. check-code Job:

• Checkout: This job begins by checking out the code from the repository.
• Uses: It uses the actions/checkout@v3 action to perform the checkout step.
• Runs on: The job runs on the ubuntu-latest runner environment provided by GitHub Actions.
• Steps:
  • Use PEP8 Action: This step installs and utilizes the PEP8 Action action to perform code linting.
    • uses:github/PEP8 Action@v1: This line specifies the SuperLinter action and its version.

2. build-and-push Job:

• Needs: This job depends on the successful completion of the check-code job (needs: check-code). This enforces code quality checks before proceeding with building and pushing images.
• Runs on: This job runs on the ubuntu-latest runner environment.
• Steps:
  1. Checkout code: Identical to the check-code job, this step retrieves the code from the repository using actions/checkout@v3.
  2. Create .env file: This step creates a .env file within the app directory. It securely stores your OpenAI API key (API_KEY) using a secret ({{ secrets.OPENAI_API_KEY }}) to prevent exposure.
  3. Set up Docker Buildx: This step leverages the docker/setup-buildx-action@v1 action to configure Docker Buildx, providing enhanced build capabilities (optional, adjust based on your needs).
  4. Setup docker-compose: This step utilizes the KengoTODA/actions-setup-docker-compose@v1.2.2 action to set up Docker Compose within the workflow. It provides access to docker-compose commands within your workflow steps. The GITHUB_TOKEN secret is required for authentication purposes.
  5. Login to DockerHub: This step securely logs in to Docker Hub using the docker/login-action@v1 action. It requires the DOCKERHUB_USERNAME environment variable (set elsewhere) and your Docker Hub token stored in the DOCKERHUB_TOKEN secret.
  6. Build and push Docker images (working directory: app):
     • Verify directory (optional): This line (ls -la) can be used for debugging purposes to confirm the working directory is set correctly.
     • Build & push using docker-compose: This primary approach utilizes docker-compose build and docker-compose push commands to build and push the images defined in your docker-compose.yml file.
     • OR tag then push individually with docker (optional): This section provides an alternative approach where you can comment out the above lines and uncomment these to tag and push each image separately using docker tag and docker push commands.
  7. Verify pushed images (optional): This step (docker images) can be used to verify the successfully pushed images to Docker Hub after the build and push process.

Getting Started:

1. Clone this repo to your local machine
2. Create a .env file in your project directory and add your OpenAI API key: OPENAI_API_KEY=your_openai_api_key.
3. Run docker-compose up -d to build and start the application.
4. Access the application in your web browser at http://localhost:8501.

Note:

• You will need an OpenAI API key to use the LLM.
• This is a basic example, and further customization might be required depending on your specific needs.

This project demonstrates a powerful approach for combining retrieval and generation tasks to create an informative and interactive user experience.