gpu-document-processing

Extract text and tables from large PDFs using GPU-accelerated workflows.

Updated Mar 17, 2026
One-click install
npx skills add https://github.com/marlo9981/Movara --skill gpu-document-processing-marlo9981
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: gpu-document-processing
Source: https://github.com/marlo9981/Movara/tree/main/examples/nvidia_deep_agent/skills/gpu-document-processing
Command: npx skills add https://github.com/marlo9981/Movara --skill gpu-document-processing-marlo9981

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Process very large PDFs and document collections efficiently by offloading heavy extraction, table parsing, and embedding generation to GPU-enabled environments while keeping agent reasoning on CPU to preserve security and cost efficiency.

Core Features & Use Cases

  • PDF Text Extraction: Layout-preserving extraction that detects headers, paragraphs, lists, tables, page numbers, and multi-column layouts.
  • Tabular Data Extraction: Convert PDF tables into CSVs or DataFrames with column type detection and support for merged cells and multi-row headers.
  • Document Chunking & Embeddings: Semantic and fixed-size chunking with overlap and GPU-accelerated embedding generation compatible with vector stores such as Milvus and ChromaDB.
  • Sandbox-as-Tool Architecture: Sends heavy processing to a GPU sandbox, enabling parallel batch processing, protecting API keys, and keeping agent state separate.
  • Batch Workflows: Designed for bulk jobs (large PDFs, 10+ documents), metadata-first processing, per-document summaries, and consolidated cross-document analysis.

Quick Start

Process the 120-page PDF annual-report-2024.pdf on the GPU sandbox to extract layout-aware text, tables, and generate embeddings for all document chunks.

Frequently Asked Questions about gpu-document-processing

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I process large PDFs with GPU acceleration for batch extraction and embeddings?

Process large PDFs by sending heavy extraction, table parsing, and embedding generation to a GPU sandbox, executing parallel batch workflows while keeping agent reasoning on CPU to preserve security and cost efficiency.

Does GPU document processing work with vector stores like Milvus and ChromaDB?

Yes, GPU document processing generates embeddings compatible with vector stores like Milvus and ChromaDB, applying semantic chunking with overlap to document chunks before storing them for downstream analysis.

How do I extract tables from multi-page PDFs and convert them to DataFrames?

Extract tables from multi-page PDFs using layout-preserving parsing that detects merged cells and multi-row headers, converting tabular PDF data into CSVs or DataFrames with column type detection.

Can I run bulk document analysis on 10+ PDFs using GPU sandbox execution?

Yes, bulk document analysis supports batch workflows for 10+ documents, utilizing GPU sandbox execution to parallelize metadata-first processing, per-document summaries, and consolidated cross-document analysis.

What is semantic chunking and how does it handle multi-column PDF layouts?

Semantic chunking divides extracted text into overlapping segments for embedding generation, while layout-preserving parsing accurately detects and processes headers, paragraphs, lists, and multi-column layouts in PDF documents.

Why use a sandbox-as-tool architecture for GPU-accelerated document processing?

A sandbox-as-tool architecture isolates heavy GPU processing from agent reasoning, protecting API keys, enabling parallel batch processing, and maintaining separate agent state for secure production document extraction workflows.