omnidoc-sdk 0.2.6

Enterprise-grade SDK for extracting text, tables, structure, and entities from documents

OmniDoc PDF Intelligence SDK

Enterprise-grade PDF understanding for Agentic AI, RAG, and automation systems

---

🚀 Overview

OmniDoc is a production-ready PDF intelligence SDK designed to convert raw PDFs (documents, slides, brochures, scanned files) into clean, structured, agent-ready outputs.

It goes beyond OCR by applying layout intelligence, semantic normalization, metric extraction, and agent-native serialization.

Built for:

• Agentic AI systems
• RAG (Retrieval-Augmented Generation)
• Knowledge Bases
• Autonomous remediation & workflows
• Enterprise document automation

---

🧠 Core Capabilities

Supported PDF Types

• Digital PDFs
• Slide decks & brochures
• Scanned PDFs
• Mixed-content PDFs

Intelligence Layers

• PDF type detection (digital vs scanned)
• OCR (Tesseract / AWS Textract)
• Layout-aware block ordering
• Vision-Language Models (Donut – optional)
• Slide → report restructuring
• Heading detection & paragraph merging
• Bullet normalization
• Table → metric extraction
• Noise & artifact removal
• RAG-ready chunking

Output Formats

• Python-native Document
• Enterprise JSON
• Agent-native TOON

---

📦 Installation

pip install omnidoc

System Dependencies (macOS / Linux)

brew install poppler tesseract

Optional:

pip install pdf2image pytesseract

---

🏗️ Architecture

PDF
 │
 ├─ Detection (Digital / Scanned)
 │
 ├─ Extraction
 │   ├─ Text
 │   ├─ OCR
 │   └─ Layout ML
 │
 ├─ Cleaning & Normalization
 │
 ├─ Slide → Report Structuring
 │
 ├─ Table & Metric Processing
 │
 └─ Output
     ├─ Document
     ├─ JSON
     └─ TOON

---

🔧 Core API

extract_pdf()

extract_pdf(
    path: str,
    enable_layout: bool = True,
    enable_cloud_ocr: bool = False,
    enable_vlm: bool = False,
    enable_pii_masking: bool = False,
    output_format: str = "document"  # document | json | toon
)

---

📄 Example — Plain Text

from omnidoc.pdf.pipeline import extract_pdf

doc = extract_pdf("sample.pdf", enable_layout=True)
print(doc.raw_text)

---

📄 Example — JSON (RAG Ready)

from omnidoc.pdf.pipeline import extract_pdf
import json

result = extract_pdf(
    "sample.pdf",
    enable_layout=True,
    enable_cloud_ocr=True,
    output_format="json"
)

print(json.dumps(result, indent=2))

---

🤖 Example — TOON (Agent Output)

toon = extract_pdf(
    "sample.pdf",
    enable_layout=True,
    output_format="toon"
)

print(toon)

---

🧪 Real Use Case — RAG Pipeline

doc = extract_pdf("strategy.pdf", output_format="json")

for chunk in doc["chunks"]:
    vector_db.add(chunk["text"], metadata=chunk)

---

🔐 Enterprise Design

• Deterministic output
• No stdout hijacking
• No numeric loss
• Agent-safe serialization
• Optional PII masking
• Cloud OCR fallback

---

📜 License

© 2025 OmniDoc — Internal / Enterprise SDK