document-extraction-module 0.1.1

Extraction / Parse of any PDF document within minutes

TODO: Update README file and add docstrings for the module. Also add instructions to build the package

📄 Document Extraction Module

A general-purpose document extraction module designed for developers. This toolkit allows you to build customizable extraction pipelines for your documents with ease using provided APIs and SDK functions.

---

🚀 Features

• Define your own extraction schema class / JSON
• Supports single and batch PDF processing
• Contains parallelization - Processes multiple PDFs and pages of each of these PDFs at a time
• Supports all English PDFs, PDFs in indic languages, scanned PDFs, PDFs with tables and complicated patterns.
• Easy LLM integration via prompt schemas

🔐 API Key Setup

If you are using GPT Models which support good extraction:export LLM_PROVIDER=open-ai and export OPENAI_API_KEY=<<your_key_here>> Specify the models you wish to use in two keys: export OPEN_AI_EXTRACTION_MODEL=<<your_extraction_model>> and export OPEN_AI_PARSE_FORMATING_MODEL=<<your_structured_output_formmating_model>> If you are using landing AI for parsing then: export VISION_AGENT_API_KEY=<<your_key_here>>

📦 Installation

First, install all the required packages using pip:

pip install -r requirements.txt

🧠 Define Your Extraction Schema

Edit extraction_class_type.py to define your custom schema. You'll use three core classes:

• ExtractionClass: Fields to extract from documents
• AIAgentClass: Prompts/instructions for the AI Agent
• OutputExampleClass: (Optional) Output examples for better LLM performance

Example class definitions:

from pydantic import BaseModel, Field
from typing import List
from prompt_processing import generate_prompt_template, generate_prompt_from_schema, PromptSchema

class ExtractionClass(BaseModel):
    name: str = Field(..., description="The name of the person.")
    age: int = Field(..., description="The age of the person.")
    hobbies: List[str] = Field(..., description="A list of hobbies.")
    address: dict = Field(..., description="The address of the person.")

class AIAgentClass(BaseModel):
    information: str = Field(..., description="Information for the AI agent.")
    instruction: str = Field(..., description="Instructions for the AI agent.")
    condition: str = Field(..., description="Conditions for the AI agent.")

class OutputExampleClass(BaseModel):
    """Optional: Provide a sample output format"""


extraction_template = generate_prompt_template(ExtractionClass)
ai_agent_template = generate_prompt_template(AIAgentClass)
output_example_template = generate_prompt_template(OutputExampleClass)

schema = PromptSchema(
    ai_agent_information=ai_agent_template,
    extract_fields=extraction_template,
    output_example=output_example_template,
)

prompt = generate_prompt_from_schema(schema.model_dump())

description in the class variable definition will act as the prompt for each of the field we wish to extract.

📂 Usage

Extracting from multiple PDFs: document_paths is an array of paths to the PDFs output_paths is whereever you would like to save the JSON files parser is True when you want to digitise the entire document, otherwise you just extract from the PDF combine_pages - Use it if you want to extract from smaller PDFs (2-3 pages) and you want to extract them together.

from extraction import extract_multiple_pdfs

document_paths = ["./152_double.pdf", "./152.pdf"]
output_paths = ["./152_extracted.json", "./152_double_extracted.json"]

multi_pdf_extraction_result = await extract_multiple_pdfs(
   input_paths=document_paths,
   output_paths=output_paths,
   parser=False,
   combine_page=False,
   prompt=prompt,
   extraction_schema_class=ExtractionClass
)

Extraction from single PDF:

from extraction import extract_multiple_pages_async
from file_process import base_64_conversion

pdf_path = "./example.pdf"
base64_images = base_64_conversation(input_type="PDF", file_path=pdf_path)

single_pdf_extraction_result = await extract_multiple_pages_async(
    input_type="PDF",
    base64_images=base64_images,
    text_inputs=[],
    combine_pages=False,
    prompt=prompt,
    extraction_schema_class=ExtractionClass
)

As of now combine_pages is disabled due to context window length related challenges.

📁 Output

The extracted information is saved as JSON in the output_path. The format matches exactly the fields defined in your ExtractionClass.

💡 Tips

• Keep prompts in AIAgentClass clear and detailed for better LLM results.
• Test your ExtractionClass with a few sample documents before scaling.
• Make use of OutputExampleClass to improve LLM consistency.