This prototype tool extracts structured data from PDF or image-based forms, converting them into interactive web forms adhering to the GOV.UK Forms schema. It targets users needing to digitize paper or scanned government forms, offering a cost-effective way to create accessible digital versions.

How It Works

The system leverages Large Language Models (LLMs), defaulting to OpenAI's GPT-4o or optionally Claude 3, to interpret form layouts. PDF documents are first rasterized into images using GraphicsMagick. These images, along with a specific prompt and JSON schema, are sent to the LLM for analysis. The LLM identifies questions, hints, field types, and even conditional logic, outputting a JSON representation of the form structure. This JSON is then used to dynamically generate multi-page web forms styled with GOV.UK Frontend components.

Quick Start & Requirements

• Install dependencies: npm install
• Install GraphicsMagick: Guide
• Set environment variable: OPENAI_API_KEY or ANTHROPIC_API_KEY
• Run locally: npm start dev
• Demo video: https://github.com/timpaul/form-extractor-prototype/assets/1590604/81580afb-e3c9-41dd-a451-be418372ef2d

Highlighted Details

• Replicates form structure in JSON following GOV.UK Forms schema.
• Generates multi-page web forms using GOV.UK Design System components.
• Supports processing of hand-drawn forms and recognizes conditional routing.
• Distinguishes between question, hint, and field text, and identifies common question types.

Maintenance & Community

• Project maintained by timpaul.
• No explicit community channels or roadmap links provided in the README.

Licensing & Compatibility

• License not specified in the README.
• Compatibility for commercial use or closed-source linking is undetermined.

Limitations & Caveats

The tool's knowledge of question types is limited, and API key input is currently restricted to environment variables, not the UI. As with many generative AI applications, outputs can be unpredictable.