Nawadiraat

Urdu and other Nastaliq scripts are notoriously hard for OCR: dense ligatures, diacritics and right-to-left flow break standard models. Off-the-shelf Tesseract Urdu (urd) produced garbled, unusable text, and there was no clean way to search or spell-check literature across Urdu, Sindhi, Farsi, Arabic and more.

Make a multi-script literature archive genuinely usable: accurate Urdu-Nastaliq OCR, a fast diacritic-insensitive dictionary and search, a real multilingual spell-checker, and a content platform covering nine provisioned locales including full RTL support.

A custom Tesseract model, urd_naw, fine-tuned from the urd base on 12,748 curated Urdu image–text pairs (RTL training), delivers markedly cleaner ligature and diacritic recognition than the baseline. It runs in a dedicated Python/FastAPI OCR microservice the Laravel app calls over HTTP. Alongside it, a from-scratch Norvig-style spell-checker (Unicode-aware, diacritic/harakat-stripping) corrects Urdu and English, and dictionary search uses MySQL REGEXP normalization for diacritic-insensitive matching.

1. 1User uploads an Urdu page image to the OCR tool
2. 2Laravel forwards it to the FastAPI OCR microservice
3. 3urd_naw (Tesseract 5) extracts Nastaliq text, rendered in Noto Nastaliq
4. 4Spell-checker normalizes and corrects Urdu/English input
5. 5Lughaat search matches diacritic-insensitively across dictionaries

• urd_naw: custom Urdu-Nastaliq Tesseract model (12,748 pairs)
• FastAPI + pytesseract OCR microservice (Dockerized)
• Norvig-style multilingual spell-checker (Unicode/RTL aware)
• Diacritic-insensitive dictionary search (MySQL REGEXP)
• Per-language feature flags (dictionary / spell-check / OCR)

• Custom Urdu-Nastaliq OCR, open-sourced to Tesseract's tessdata_contrib
• Multilingual spell-checker for Urdu & English
• Lughaat: multiple dictionaries with rich lexicography
• Literature library: poets, poetry, prose, kids' stories, recitations
• Nine provisioned locales with full RTL support
• Filament admin with bulk CSV/XLSX import pipelines

A Laravel 10 monolith (Blade + Livewire + Alpine, Filament 3 admin, MySQL) handles the site, content and dictionaries. OCR is isolated in a standalone Python FastAPI microservice running Tesseract 5 with the bundled urd_naw model, containerized with Docker and called over HTTP with a health check. Locale, RTL and per-language feature availability are all data-driven.

A fast, RTL-aware multilingual front end (literature browsing, Lughaat, spell-checker and OCR tools) with Nastaliq typography, backed by a Filament admin where the team bulk-imports words, poems and poets via CSV/XLSX.

• FastAPI OCR microservice (internal HTTP)
• YouTube (recitations & video)
• Google Analytics 4
• Noto Nastaliq Urdu & Abdo Line fonts
• Partners: Govt of Punjab & Sindh, PILAC, Sindhi Language Authority

The Laravel app and the Dockerized OCR service are deployed as separate containers (OCR on its own port with CORS locked to the domain and a /health check), so OCR scales independently of the website. The urd_naw model is published openly in Tesseract's tessdata_contrib repository.