Scanned Urdu is messy
ChallengeOCR on scanned Urdu pages produces many errors.
How we solved itGoogle Cloud Vision with an Urdu language hint, followed by an AI proofreading pass.
A non-profit AI chatbot over the works of Syed Abul A’la Maududi — Urdu books digitised with OCR and AI proofreading, then opened up to questions in Urdu and English.
Maududi’s books, articles and Tafheem-ul-Quran are spread across websites and scanned Urdu PDFs. Scanned Urdu needs careful correction before anyone can search it — let alone ask it questions.
First the library became clean, structured Urdu text; then readers could talk to it.
Before any chatbot, the library had to become clean, searchable Urdu text.
6 core features
The pipeline behind the chatbot.
Books and articles gathered into one corpus.
Google Cloud Vision reads the Urdu pages.
Gemini fixes OCR errors, keeping the text intact.
Unicode text, linked JSON and Word volumes.
Readers question the library in Urdu or English.
A Python pipeline turns the sources into clean Urdu text; a React chat app streams answers from an API built on that corpus.
Chat appReact · Urdu & English
Urdu OCRGoogle Cloud Vision
ProofreadingGemini
PipelinePython · python-docxChallengeOCR on scanned Urdu pages produces many errors.
How we solved itGoogle Cloud Vision with an Urdu language hint, followed by an AI proofreading pass.
ChallengeA scholar’s text must never be rewritten.
How we solved itThe proofreading step is instructed to fix OCR errors only and preserve the full content.
ChallengeUrdu needs a right-to-left layout and Nastaliq type.
How we solved itRight-to-left documents, and a chat that switches direction and font for Urdu.
ChallengeBooks, articles and Tafheem live in different places.
How we solved itConverters bring them into one structured, linked corpus.
Proven, well-supported tools — chosen so the platform is easy to run and to extend.
PythonPipelineGoogle CloudUrdu OCR (Vision)Google GeminiAI proofreadingReactChat appPlatforms like Ask Maududi can now do far more for the people who use them. These are the upgrades we’d add today, using the AI and automation we build for clients.
Every answer linked to its book and page.
Ask by voice and listen to the answer.
Newer vision models for difficult scans.
Test answers against reviews by scholars.
The whole library in your pocket.
Let other assistants search the library.
Also in production: we build AI over books and archives — see Khurshid Answers and Pakistan LawBot.
Tell us who will use it and what it must do. We’ll map the first release, the stack and the right team — in a free 30-minute strategy call.
