LLM-Powered Invoice & Receipt Extractor (OSS)

Struggling to get real receipt/invoice data for your AI models? I built an open-source generator using LLMs (JSON output, no templates)

Link: https://github.com/WellApp-ai/Well/tree/main/ai-receipt-generator

Sample output: https://imgur.com/a/YtFSodj

When you're building AI systems to extract structured data from receipts, invoices, and other financial docs, there's one big bottleneck: Realistic, diverse, high-volume training data.

Most open datasets are:

• Too clean (template-generated)
• Too uniform (Western formats only)
• Not legally usable at scale

So I built this little open-source tool that uses LLMs to generate synthetic receipts in JSON format, fully customizable via prompt + config. No PDFs, no OCR simulation — just structured text output designed for evals, testing, or fine-tuning.

Key features:

• Works with OpenAI, local models, Claude, etc. (LLM-agnostic)
• JSON schema for receipts/invoices, easy to customize
• Faker fallback if you don’t want to hit a model
• Locale-aware: useful for global format simulation
• Configurable weirdness: broken totals, missing fields, typos, etc.

This helped us stress-test our document parser with realistic, non-trivial edge cases that templates couldn’t replicate.

Curious if anyone else here is:

• Generating synthetic data for document AI
• Testing LLM-based extractors or OCR+LLM combos
• Building eval suites for financial AI models

Would love feedback, ideas, or thoughts on how you’d extend this.