Read
See the script
OCR for printed and handwritten Pashto, Urdu, and Latin text, from a single line to a full document.
Zirak AI · Peshawar
Intelligence in your mother tongue.
Zirak Technologies builds machines that read, write, listen, speak, and understand low-resource languages of South Asia and the Middle East — Pashto, Dari, Urdu, Persian, and Arabic — with the same care the industry gives to English.
1MPsOCR training images
10KPublic OCR benchmark
11Published papers
2024Founded in Peshawar
01 — Mission
No language left behind.
Cutting-edge models still fail where the script is cursive, the data is scarce, and the speakers number in the tens of millions. Zirak AI closes that gap with optical character recognition, speech, and language systems designed around local linguistic needs — preserving diversity while opening the digital world.
02 — Capabilities
Five senses for language.
Write
Shape the words
Segmentation, spelling, part-of-speech, and translation models that respect how these languages are actually written.
Listen
Hear the voice
Pashto speech-to-text for transcription, dictation, and voice interfaces, including native accents.
Speak
Answer aloud
Text-to-speech tuned for clarity, tone, and regional Pashto dialects, ready for apps and platforms.
Understand
Know the meaning
Monolingual Pashto BERT, offensive-language detection, and benchmarks the research community can build on.
NLPashto
An open toolkit for spelling, segmentation, tagging, and moderation in a language spoken by over 50 million people.
03 — Flagship
PsOCR, the first million-image Pashto benchmark.
One million synthetic images, a thousand font families, and a public 10,000-image test set. We used it to measure how today’s large multimodal models — Gemini, GPT-4o, Claude, Grok, Qwen, and others — actually read Pashto.
Contact
Build with us.
Governments, universities, and product teams use Zirak AI when the language on the page is not English.