xiry.

AI · Web app

PDF to Structured Data

Extract structured JSON from PDFs using natural language instructions and Gemini

From the Xiry library · Updated 10 October 2026

Visit huggingface.co ↗

This open-source web application extracts structured JSON data from uploaded PDFs using Google's Gemini 2.0 Flash model. Users describe what they want to pull out in plain text, and the system automatically generates an extraction schema and returns clean JSON. It removes the need for custom parsing scripts or fixed templates.

What it's for

  • Extract specific fields from scanned invoices and receipts into JSON format
  • Parse handwritten or unstructured government forms without writing regex rules
  • Convert legacy PDF documents into standardized schema-based records

Features

  • Natural language prompting to define required extraction fields
  • Automatic schema generation based on user instructions
  • Gemini 2.0 Flash integration for fast multimodal document processing
  • Direct JSON output export
  • Open-source Next.js stack deployable via Docker or locally

Pricing

Free Hugging Face Space; requires a Gemini API key if self-hosted

Details

Type
Web app
Category
AI
Topic
Document Processing
Open source
Yes
Platforms
Web, Self-hosted
For
Developers, data engineers, and researchers handling unstructured PDF documents
Alternative to
Textract, Docsumo, LlamaParse
Tags
pdf, json, gemini, data extraction, multimodal, huggingface

Why it was saved

  • @_philschmid, 27 February 2025: Turns messy or handwritten PDFs (forms, invoices) into clean structured data without building a custom parser. The schema view shows how the tool maps your description to fields like formNumber, formStartDate, liabilityStartDate and liabilityEndDate.

Related

Updated