Remixonwin/amt-airframe-handbook-dataset

AMT Airframe Handbook Dataset

A comprehensive dataset extracted from the FAA Aviation Maintenance Technician (AMT) Airframe Handbook, containing text content and rendered page images suitable for training vision-language models.


Overview

This dataset was created using the doc-parser-engine - a production-grade document parsing engine with HuggingFace integration. The source document is the FAA Aviation Maintenance Technician Airframe Handbook, a foundational text for aircraft maintenance professionals.

Dataset Statistics

Metric	Value
Total Pages	252
Word Count	80,740
Images	253 (rendered pages)
Paragraphs	5,521
Sections	7
Chapters	1

Data Schema

The dataset contains 19 columns with the following structure:

Document Metadata

• doc_id (string): Unique document identifier
• title (string): Document title
• authors (list[string]): Document authors
• doc_type (string): Document type (e.g., "pdf")
• source_path (string): Original file path
• created_at (string): Creation timestamp
• parsed_at (string): Parsing timestamp

Content Statistics

• word_count (int32): Total word count
• page_count (int32): Total page count
• image_count (int32): Number of images extracted
• table_count (int32): Number of tables extracted

Structural Content

• chapters (list[string]): Chapter titles
• sections (struct): Section details including:
  • section_id: Section identifier
  • title: Section title
  • level: Section hierarchy level
  • content: Section content
  • semantic_label: Semantic classification
  • page_start: Starting page number
  • word_count: Section word count

Paragraphs

• paragraphs (struct): Paragraph details including:
  • para_id: Paragraph identifier (e.g., "para_00001")
  • text: Paragraph content
  • section_id: Parent section
  • page: Page number
  • word_count: Paragraph word count

Tables

• tables (struct): Table details including:
  • table_id: Table identifier
  • headers: Table headers
  • num_rows: Number of rows
  • num_cols: Number of columns
  • caption: Table caption

Images

• images (struct): Image details including:
  • image_id: Unique image identifier
  • caption: Image caption (when available)
  • alt_text: Alternative text
  • ocr_text: OCR extracted text
  • width: Image width in pixels
  • height: Image height in pixels
  • format: Image format (PNG)
  • page: Source page number
  • file_path: File path reference
  • category: Image category

Additional Metadata

• metadata (string): Additional metadata
• footnotes_count (int32): Number of footnotes
• references_count (int32): Number of references

Usage Example

Loading the Dataset

from datasets import load_dataset

# Load the dataset from HuggingFace
dataset = load_dataset("Remixonwin/amt-airframe-handbook-dataset")

# Access the training split
train_data = dataset["train"]

# Get the first (and only) document
doc = train_data[0]

print(f"Title: {doc['title']}")
print(f"Word Count: {doc['word_count']}")
print(f"Page Count: {doc['page_count']}")
print(f"Image Count: {doc['image_count']}")

Accessing Paragraphs

# Access structured paragraphs
paragraphs = doc["paragraphs"]
print(f"Total paragraphs: {len(paragraphs['para_id'])}")

# Get first paragraph
first_para = {
    "id": paragraphs["para_id"][0],
    "text": paragraphs["text"][0],
    "page": paragraphs["page"][0],
    "word_count": paragraphs["word_count"][0]
}
print(first_para)

Accessing Images

# Access images with metadata
images = doc["images"]
print(f"Total images: {len(images['image_id'])}")

# Get first image details
first_image = {
    "id": images["image_id"][0],
    "width": images["width"][0],
    "height": images["height"][0],
    "page": images["page"][0],
    "format": images["format"][0]
}
print(first_image)

Image Extraction Details

This dataset uses page rendering to capture vector graphics and diagrams from the PDF. The original PDF contains technical drawings, diagrams, and figures rendered as vector graphics which are not extracted by traditional image extraction methods.

• Rendering DPI: 150
• Image Format: PNG
• Page Images: Full-page renders at 1242x1755 pixels

Use Cases

This dataset is ideal for:

1. Vision-Language Model Training: Train multi-modal models on aviation maintenance content
2. Document Understanding: Research on technical document parsing and understanding
3. Question Answering: Build Q&A systems for aviation maintenance training
4. Information Retrieval: Create search systems for technical documentation
5. Content Analysis: Analyze structure and content of technical handbooks

Dataset Creation

The dataset was created using the doc-parser-engine with the following configuration:

from doc_parser_engine.core import DocParserEngine

engine = DocParserEngine(
    enable_captioning=True,
    caption_model="api",
    llm_api_base="http://0.0.0.0:7543",
    llm_model="gpt-4o",
    render_pages=True,    # Enable page rendering
    render_dpi=150       # 150 DPI for rendered images
)

doc = engine.parse("amt_airframe.pdf")

# Create HuggingFace dataset
dataset = engine.to_hf_dataset(
    [doc],
    schema="full",
    include_images=True,
    push_to_hub=True,
    hub_repo="Remixonwin/amt-airframe-handbook-dataset"
)

License

This dataset contains content from the FAA Aviation Maintenance Technician Handbooks. Please verify the appropriate usage rights for your intended application.

Citation

If you use this dataset in your research, please cite:

@dataset{amt_airframe_handbook_2026,
  author = {Federal Aviation Administration},
  title = {Aviation Maintenance Technician Airframe Handbook},
  year = {2026},
  publisher = {HuggingFace},
  dataset_url = {https://huggingface.co/datasets/Remixonwin/amt-airframe-handbook-dataset}
}

Related Datasets

• AMT Powerplant Handbook Dataset - Complementary dataset covering aircraft powerplant systems

Contact

For questions or issues with the dataset, please open an issue on the GitHub repository.

---

Dataset last updated: February 2026