siraaj-dot-ocr-service / docs/README.md
Documentation
Last updated: 4/16/2026GitHub
Documentation
Service Documentation
- Architecture & Flows — processing flows, fallback strategies, prompts
- Migration Complete — Triton migration checklist
Research
- OCR Model Comparison Report — benchmark across 20+ models (English + Arabic)
- Hard English Benchmark — 17 models on 9 difficult pages with ground truth scoring
- Benchmark Suite — scripts, baselines, and raw results
- Router Benchmark — SigLIP2 zero-shot language router for EN/AR page classification
- Triton Concurrency Tuning — instance count optimization for router and OCR pipeline on L40S