Overview

OCR Lab is a high-throughput local document intelligence and hybrid presentation OCR engine powered by Datalab’s Indra OCR (~1M parameter ONNX Runtime model).

It runs 100% locally on CPU with zero GPU requirement, unlimited batch file processing, and persistent SQLite storage.

Key Highlights

  • Local & Private: Runs entirely on local CPU with ONNX Runtime. No cloud APIs, no external telemetry, zero data leakage.
  • Hybrid Native + Vision Extraction: Fast digital text extraction with selective vision OCR fallback for complex diagram/slide layouts.
  • Slide & Presentation Intelligence: Handles presentation decks, code snippets, architecture diagrams, and multi-column tables.
  • Batch Processing Engine: Multi-file queue with progress tracking, stage metrics, and instant preview.
  • Persistent Storage: SQLite WAL with content-addressable hash deduplication.

Architecture

  • Model: Datalab Indra OCR (~1M params, ONNX CPU)
  • Backend: Python 3.12+, FastAPI, SQLite WAL
  • Deployment: Docker & Docker Compose ready