Showcase: Real-Time Invoice Parser using Qwen2.5-VL
🚀 I just built a "Zero-Cost" AI Invoice Parser 📑🤖
Most Document AI solutions come with a heavy price tag and privacy concerns. I wanted to see if I could build a professional-grade alternative using open-source Vision-Language Models (VLMs) and ZeroGPU infrastructure.
The result: IntelliReceipt.
How it works:
- Intelligent Routing: Instead of forcing the AI to guess, I implemented a Validation Gate. The system first identifies if the image is actually a document; if it’s a random photo, it rejects it to prevent "AI hallucinations."
- Dual-Stage Processing: A "Router" identifies the vendor, and a "Specialist" then applies a precise JSON schema to extract data with 100% structural accuracy.
- Global Layout Recognition: Built as an international tool, the system excels at interpreting diverse fiscal layouts and currencies, handling complex horizontal writing flows that often trip up traditional OCR engines.
Key Features:
✅ PDF & Image Support: Automatically converts PDF pages for visual processing.
✅ Real-Time Feedback: Includes a custom progress meter and document preview.
✅ Privacy-First: Runs on local/private environments with no third-party API calls.
Check out the live Space here: https://huggingface.co/spaces/Liviu16/InvoiceRecon
IntelliReceipt handles Romanian RON invoices specifically well. It adds that local expertise touch that many global tools struggle with!
#AI #MachineLearning #Python #OpenSource #DocumentAI #Innovation #HuggingFace
