Build a containerized PDF to Markdown converter using Docling and Streamlit ☁️

Introduction

In one of my last discussions, I was asked to help a business partner to start using Docling for document preparation to be sent to a RAG. Putting aside the project details, the application is meant to run on a Kubernetes cluster.

The sample provided here provides the steps to create a Python application using Docling document conversion capabilities, and using Streamlit to build the web UI. The application can be containerized to be deployed on a k8s cluster.

What is Docling

Docling simplifies document processing, parsing diverse formats — including advanced PDF understanding — and providing seamless integrations with the gen AI ecosystem.

Features

🗂️ Parsing of multiple document formats incl. PDF, DOCX, XLSX, HTML, images, and more
📑 Advanced PDF understanding incl. page layout, reading order, table structure, code, formulas, image classification, and more
🧬 Unified, expressive DoclingDocument representation format
↪️ Various export formats and options, including Markdown, HTML, and lossless JSON
🔒 Local execution capabilities for sensitive data and air-gapped environments
🤖 Plug-and-play integrations incl. LangChain, LlamaIndex, Crew AI & Haystack for agentic AI
🔍 Extensive OCR support for scanned PDFs and images
💻 Simple and convenient CLI

The sample application

Disclaimer: this is a very simple implementation, in order to showcase the capacities. It should be enhanced for a scaled / industrialized application.

On Intel machines, most likely there is a need to troubleshoot the configuration in order to make it runnable. The Conda activation will probably be required. Hereafter is what I had to do on my laptop.

# conda installation
v=3; f=Miniconda${v}-latest-MacOSX-x86_64.sh; cd $TMPDIR; { curl -LfOsS https://repo.anaconda.com/miniconda/$f ; cd -; } && bash $TMPDIR/$f -b && echo ". ~/miniconda${v}/etc/profile.d/conda.sh" >> ~/.bash_profile; . ~/miniconda${v}/bin/activateconda

# initialization
conda init


source ~/.bashrc
conda create --name py11 python==3.11  

# activation
conda activate py11

And hereafter the application;

import streamlit as st
import os

from pathlib import Path

from docling.backend.docling_parse_backend import DoclingParseDocumentBackend
from docling.datamodel.base_models import InputFormat
from docling.datamodel.pipeline_options import (
    EasyOcrOptions,
    OcrMacOptions,
    PdfPipelineOptions,
    RapidOcrOptions,
    TesseractCliOcrOptions,
    TesseractOcrOptions,
)
from docling.document_converter import DocumentConverter, PdfFormatOption

def main():
   def main():
    st.title("File Selection App")

    # Initialize session state for storing the concatenated path
    if 'full_file_path' not in st.session_state:
        st.session_state.full_file_path = None

    # Add description
    st.write("Select a file from your computer to view its path information")

    # File uploader with type filtering
    allowed_types = ["pdf", "doc", "docx", "ppt", "pptx", "jpg", "jpeg"]
    uploaded_file = st.file_uploader(
        "Choose a file",
        type=allowed_types,
        help="Supported files: PDF, Word, PowerPoint, and JPEG"
    )

    # Display file details if a file is uploaded
    if uploaded_file is not None:
        # Get the current working directory
        current_dir = os.getcwd()

        # Create full path (this will be where Streamlit temporarily stores the file)
        full_path = os.path.join(current_dir, uploaded_file.name)

        # Store the concatenated path in session state
        st.session_state.full_file_path = full_path

        file_details = {
            "Filename": uploaded_file.name,
            "Full Path": full_path,
            "File size": f"{uploaded_file.size / 1024:.2f} KB",
            "File type": uploaded_file.type,
            "Concatenated Path Variable": st.session_state.full_file_path
        }

        st.write("### File Details:")
        for key, value in file_details.items():
            st.write(f"**{key}:** {value}")

        # Example of using the concatenated path in the app
        st.write("### Using the Concatenated Path")
        st.code(f"""
        # You can access the full file path anywhere in your app using:
        full_file_path = st.session_state.full_file_path

        # Example usage:
        if st.session_state.full_file_path:
            # Do something with the path
            print(f'Working with file: {st.session_state.full_file_path}')
        """)

        # Add a button to demonstrate using the path
        if st.button("Print Path to Console"):
            st.write(f"Path has been printed to console: {st.session_state.full_file_path}")
            print(f"Full file path: {st.session_state.full_file_path}")

        # Add a note about the path
        st.info("Note: The full path shown is where Streamlit temporarily stores the uploaded file. " 
                "To get the original file path, you would need to use a different approach as browsers "
                "don't provide the original file path for security reasons.")

    ######        
    ##
    input_doc = st.session_state.full_file_path
    ##

    pipeline_options = PdfPipelineOptions()
    pipeline_options.do_ocr = True
    pipeline_options.do_table_structure = True
    pipeline_options.table_structure_options.do_cell_matching = True

    # Any of the OCR options can be used:EasyOcrOptions, TesseractOcrOptions, TesseractCliOcrOptions, OcrMacOptions(Mac only), RapidOcrOptions
    # ocr_options = EasyOcrOptions(force_full_page_ocr=True)
    # ocr_options = TesseractOcrOptions(force_full_page_ocr=True)
    # ocr_options = OcrMacOptions(force_full_page_ocr=True)
    # ocr_options = RapidOcrOptions(force_full_page_ocr=True)
    ocr_options = TesseractCliOcrOptions(force_full_page_ocr=True)
    pipeline_options.ocr_options = ocr_options

    converter = DocumentConverter(
        format_options={
            InputFormat.PDF: PdfFormatOption(
                pipeline_options=pipeline_options,
            )
        }
    )

    doc = converter.convert(input_doc).document
    md = doc.export_to_markdown()
    print(md)

if __name__ == "__main__":
    main()

In order to make the application distributable I set up the following “requirements.txt” (not the best type of requirements which could be written by a real Python developer 😅).

streamlit
pathlib
docling
logging
time
datetime
pandas
tensorflow

To test the app locally, the following simple command line does the job!

streamlit run app.py

And there goes the Dockerfile.

# Use an official Python runtime as a parent image
FROM python:3.11-slim

# Set environment variables
ENV PYTHONUNBUFFERED=1 \
    PYTHONDONTWRITEBYTECODE=1 \
    PIP_NO_CACHE_DIR=1

# Set the working directory in the container
WORKDIR /app

# Install system dependencies
RUN apt-get update && apt-get install -y \
    build-essential \
    curl \
    software-properties-common \
    git \
    && rm -rf /var/lib/apt/lists/*

# Copy only requirements first to leverage Docker cache
COPY requirements.txt .

# Install Python dependencies
RUN pip install --upgrade pip setuptools wheel \
    && pip install -r requirements.txt \
    && rm -rf /root/.cache/pip

# Copy the rest of the application
COPY . .

# Expose Streamlit port
EXPOSE 8501

# Add healthcheck
HEALTHCHECK CMD curl --fail http://localhost:8501/_stcore/health

# Set Streamlit configuration
ENV STREAMLIT_SERVER_PORT=8501
ENV STREAMLIT_SERVER_ADDRESS=0.0.0.0

# Run the application
CMD ["streamlit", "run", "app.py"]

I use Podman to build and deploy the image, by aliasing Podman to Docker, all Docker commands could be used.

docker build -t app:latest .
#
docker run -p 8501:8501 app:latest
# or 
docker run -d -p 8501:8501 app:latest

Conclusion

This article shows the ease of Docling implementation in order to make enterprise documents ready to be ingested easily for a RAG application for example.

Thanks for reading 🙏

Useful links

• Docling repository: https://github.com/DS4SD/docling
• Docling documentation and samples: https://ds4sd.github.io/docling/
• Docling PDF to Markdown sample: https://ds4sd.github.io/docling/examples/full_page_ocr/