PDF Q&A Example#

This example demonstrates how to use NVIDIA’s Context-Aware RAG (CA-RAG) system for PDF document processing and question answering. The example shows how to:

  • Extract text and tables from PDF documents

  • Ingest documents into the CA-RAG system

  • Perform question answering on the ingested documents

Prerequisites#

  1. NVIDIA API Keys:

    • Get your API keys from: build.nvidia.com

    • Export the following environment variables:

      export NVIDIA_BUILD_API_KEY=your_build_api_key
export NVIDIA_API_KEY=your_api_key

  2. Install NV-Ingest:

    • Follow the installation instructions at: NV Ingest

Setup#

  1. Start the required services using docker-compose:

  2. Start the NV-Ingest client:

    • The example uses a pipeline configuration with Milvus for vector storage

Usage#

  1. Document Processing:

    • Place your PDF documents in the data/ directory

    • The example processes PDFs to extract:

      • Text content

      • Tables

      • Charts

      • Note: Image extraction is not supported in the current version

  2. Document Ingestion:

    • Documents are processed and uploaded to the ingestion service

    • The system maintains document order and metadata

    • A terminating document is added to mark the end of the document set

  3. Question Answering:

    • Initialize the retrieval service with the same UUID used for ingestion

    • Send questions to the retrieval service

    • Receive answers based on the ingested document content

Example Notebook#

The pdf_qna.ipynb notebook provides a step-by-step walkthrough of the entire process, including:

  • Service initialization

  • Document processing and ingestion

  • Question answering examples

Notes#

  • The current version does not support image extraction from PDFs

  • For optimal PDF processing, especially for scanned documents, consider using the nemoretriever_parse extraction method

  • The system uses markdown format for text extraction output