""" DocuMint AI - Invoice & Document Intelligence API Deploy on Hugging Face Spaces (Gradio) for free hosting + automatic API endpoint. """ import os import json import time import threading import tempfile import gradio as gr from PIL import Image try: import spaces gpu_decorator = spaces.GPU(duration=60) except (ImportError, Exception): gpu_decorator = lambda f: f from extraction import extract_invoice, extract_from_pdf, ExtractionResult # --------------------------------------------------------------------------- # Configuration # --------------------------------------------------------------------------- HF_TOKEN = os.environ.get("HF_TOKEN", "") APP_TITLE = "DocuMint AI" APP_DESCRIPTION = """ ## AI-Powered Invoice & Document Extraction Upload an invoice, receipt, or purchase order and get **structured JSON data** instantly. **How it works:** 1. Upload an image (PNG, JPG) or PDF of your invoice 2. Our AI extracts vendor details, line items, amounts, dates, and tax info 3. Download the structured JSON or integrate via our API **Extraction Methods:** - 🧠 **AI Vision** (Primary): Uses Qwen2.5-VL for high-accuracy document understanding - 📝 **OCR + Patterns** (Fallback): Tesseract OCR with intelligent pattern matching --- """ # --------------------------------------------------------------------------- # Processing Functions # --------------------------------------------------------------------------- @gpu_decorator def process_document(file, use_ocr_only: bool = False): """Process an uploaded document and return extraction results.""" if file is None: return "⚠️ Please upload a document first.", "{}" start_time = time.time() try: file_path = file if isinstance(file, str) else file.name file_ext = os.path.splitext(file_path)[1].lower() if file_ext == ".pdf": with open(file_path, "rb") as f: pdf_bytes = f.read() results = extract_from_pdf( pdf_bytes, hf_token=HF_TOKEN, force_ocr=use_ocr_only, ) if not results: return "⚠️ Could not extract data from PDF.", "{}" # Merge multi-page results or return first page result = results[0] if len(results) > 1: # For multi-page invoices, merge line items for r in results[1:]: result.line_items.extend(r.line_items) else: image = Image.open(file_path) result = extract_invoice( image, hf_token=HF_TOKEN, force_ocr=use_ocr_only, ) total_time = int((time.time() - start_time) * 1000) result.processing_time_ms = total_time # Check for informative failure modes if result.extraction_method.startswith("failed"): method = result.extraction_method if "no_token" in method: warning = ( "⚠️ **Extraction failed:** OCR could not read this document " "(likely handwritten or low quality).\n\n" "**AI Vision is not enabled.** Set the `HF_TOKEN` Space Secret " "to unlock Qwen2.5-VL, which handles handwritten text, " "non-English documents, and complex layouts with much higher accuracy." ) elif "vlm" in method: warning = ( "⚠️ **Extraction failed:** AI Vision models could not process this document, " "and OCR could not extract readable text either.\n\n" f"**Debug info:** `{method}`\n\n" "This can happen when HF Inference API models are loading or rate-limited. " "Try again in a minute, or check that your `HF_TOKEN` has valid API access." ) else: warning = ( "⚠️ **Extraction failed:** Could not extract readable text from this document. " "The image may be too blurry, rotated, or handwritten for OCR.\n\n" "Try uploading a clearer scan, or uncheck 'OCR only' to use AI Vision." ) json_output = json.dumps(result.to_dict(), indent=2, default=str) return warning, json_output # Format summary summary = format_summary(result) json_output = json.dumps(result.to_dict(), indent=2, default=str) return summary, json_output except Exception as e: return f"❌ Error: {str(e)}", "{}" def format_summary(result: ExtractionResult) -> str: """Format extraction result as a readable summary.""" lines = [] lines.append(f"### Extraction Results") lines.append(f"**Method:** {result.extraction_method}") lines.append(f"**Confidence:** {result.confidence:.0%}") lines.append(f"**Processing Time:** {result.processing_time_ms}ms") lines.append("") if result.vendor_name: lines.append(f"**Vendor:** {result.vendor_name}") if result.invoice_number: lines.append(f"**Invoice #:** {result.invoice_number}") if result.invoice_date: lines.append(f"**Date:** {result.invoice_date}") if result.due_date: lines.append(f"**Due Date:** {result.due_date}") if result.currency: lines.append(f"**Currency:** {result.currency}") lines.append("") if result.line_items: lines.append("**Line Items:**") lines.append("") lines.append("| Description | Qty | Unit Price | Amount |") lines.append("|---|---|---|---|") for item in result.line_items: desc = item.description if isinstance(item, object) and hasattr(item, 'description') else item.get('description', '') qty = item.quantity if hasattr(item, 'quantity') else item.get('quantity', '') price = item.unit_price if hasattr(item, 'unit_price') else item.get('unit_price', '') amt = item.amount if hasattr(item, 'amount') else item.get('amount', '') lines.append(f"| {desc} | {qty or '-'} | {price or '-'} | {amt or '-'} |") lines.append("") if result.subtotal is not None: lines.append(f"**Subtotal:** {result.currency} {result.subtotal:,.2f}") if result.tax_amount is not None: tax_str = f"**Tax:** {result.currency} {result.tax_amount:,.2f}" if result.tax_rate: tax_str += f" ({result.tax_rate})" lines.append(tax_str) if result.total_amount is not None: lines.append(f"**Total:** {result.currency} {result.total_amount:,.2f}") if result.customer_name: lines.append(f"\n**Billed To:** {result.customer_name}") return "\n".join(lines) # --------------------------------------------------------------------------- # Gradio Interface # --------------------------------------------------------------------------- def create_demo(): """Create the Gradio demo interface.""" with gr.Blocks( title=APP_TITLE, theme=gr.themes.Soft( primary_hue="blue", secondary_hue="slate", ), css=""" .main-header { text-align: center; margin-bottom: 1rem; } .main-header h1 { font-size: 2rem; color: #1e40af; } .output-json { font-family: monospace; font-size: 0.85rem; } footer { display: none !important; } """ ) as demo: gr.HTML("""

📄 DocuMint AI

Extract structured data from invoices, receipts & purchase orders

""") gr.Markdown(APP_DESCRIPTION) with gr.Row(): with gr.Column(scale=1): file_input = gr.File( label="Upload Invoice (Image or PDF)", file_types=["image", ".pdf"], type="filepath", ) ocr_only = gr.Checkbox( label="Use OCR only (skip AI vision model)", value=False, info="Check this if the AI model is unavailable or you want faster processing", ) extract_btn = gr.Button( "🔍 Extract Data", variant="primary", size="lg", ) with gr.Column(scale=1): summary_output = gr.Markdown( label="Extraction Summary", value="*Upload a document and click Extract to begin.*", ) with gr.Row(): json_output = gr.Code( label="Structured JSON Output", language="json", lines=20, ) # API usage instructions with gr.Accordion("🔌 API Usage", open=False): gr.Markdown(""" ### Use DocuMint AI as an API This Gradio app is automatically exposed as an API endpoint on Hugging Face Spaces. **Python Example:** ```python from gradio_client import Client client = Client("YOUR_SPACE_URL") result = client.predict( file="path/to/invoice.png", use_ocr_only=False, api_name="/predict" ) summary, json_data = result print(json_data) ``` **cURL Example:** ```bash curl -X POST "YOUR_SPACE_URL/api/predict" \\ -H "Content-Type: multipart/form-data" \\ -F "file=@invoice.png" \\ -F "use_ocr_only=false" ``` **Response Format:** ```json { "vendor_name": "Acme Corp", "invoice_number": "INV-2024-001", "invoice_date": "2024-01-15", "total_amount": 1250.00, "currency": "USD", "line_items": [...], "tax_amount": 225.00, "confidence": 0.92 } ``` """) # Pricing section with gr.Accordion("💰 Pricing Plans", open=False): gr.Markdown(""" | Plan | Pages/Month | Price | Per Page | |------|------------|-------|----------| | **Free** | 50 | $0 | $0.00 | | **Starter** | 500 | $29/mo | $0.058 | | **Pro** | 2,000 | $79/mo | $0.040 | | **Business** | 10,000 | $199/mo | $0.020 | | **Enterprise** | Unlimited | Custom | Contact us | *All plans include JSON + CSV export, API access, and email support.* *Pro and above include priority processing and webhook notifications.* 📧 Contact: **documint.ai@gmail.com** for enterprise pricing and custom integrations. """) extract_btn.click( fn=process_document, inputs=[file_input, ocr_only], outputs=[summary_output, json_output], ) return demo # --------------------------------------------------------------------------- # Entry Point # --------------------------------------------------------------------------- if __name__ == "__main__": demo = create_demo() # Launch Gradio with prevent_thread_lock so we can add custom API routes # to Gradio's internal FastAPI app. This preserves ZeroGPU compatibility # (demo.launch handles @spaces.GPU registration) while also serving our # REST endpoints for the React frontend. app, local_url, _ = demo.launch( server_name="0.0.0.0", server_port=7860, share=False, ssr_mode=False, prevent_thread_lock=True, ) # Inject our REST API routes into Gradio's FastAPI app. # Gradio's app already includes CORSMiddleware (allow_origins=["*"]), # so our routes inherit CORS handling — no extra middleware needed. from api import api_router from api_runs import runs_router from api_prompts import prompts_router app.include_router(api_router) app.include_router(runs_router) app.include_router(prompts_router) print(f"✅ REST API routes mounted at {local_url}api/v1/") # Keep the main thread alive (uvicorn runs in a background thread) threading.Event().wait()