We just launched GPU-powered OCR (100+ languages) in Docpose.cloud API

Hey IndieHackers 👋

We’ve rolled out integrated, GPU-powered OCR inside the Docpose.cloud conversion API for paid users.

Now when a scanned PDF or image is uploaded, our system automatically detects non-searchable content and applies OCR directly inside the conversion pipeline — no extra endpoint, no additional workflow.

What’s included:

• GPU-accelerated OCR processing
• 100+ languages supported
• Automatic language detection
• Optional manual language override
• Table detection with structured layout preservation
• Handles complex layouts (multi-column docs, forms, mixed content)
• Works with scanned PDFs and images (JPG, PNG, TIFF, etc.)
• Output includes fully searchable PDFs

OCR-enabled conversions support output to:

TXT, RTF, DOCX, XLSX, XLS, CSV, PPTX, searchable PDF, EPUB, DJVU, FB2 — and many more formats supported by our pipeline.

Examples:

Scanned invoice → structured XLSX
Scanned contract → editable DOCX
Scanned legal agreement → searchable PDF
Scanned report → searchable PDF with selectable text
Scanned book → EPUB
Image → clean TXT or CSV

We focused heavily on preserving tables, formatting, and structured data because real-world documents aren’t simple paragraphs — they’re invoices, financial reports, compliance files, and legal documents where layout matters.

Architecture-wise:
OCR runs inside our existing S3-based job pipeline. No local persistence. Source files can be deleted instantly. Converted outputs follow our standard 24-hour retention policy. Logs remain scoped to system-level analytics.

This is built for SaaS teams handling document ingestion, legal tech workflows, compliance systems, financial automation, and archive digitization — where speed, structure, and accuracy are critical.

If you're running OCR in production, what’s been your biggest bottleneck: accuracy, layout preservation, or scaling under load?