PESHAWAR: In an inspiring story of young talent and quiet determination, two recent graduates from the Department of Computer Science, University of Peshawar, have independently created advanced artificial intelligence tools for Pashto – a language spoken by over 60 million people worldwide but long neglected in global technology.
Junaid Ahmed developed the first LLM (large language model) built specifically for Pashto, focusing on the Peshawari dialect. Working alone without any funding or support, he trained it on 3.4 million Pashto documents and over 126,000 instruction pairs.
Experts see these achievements as a turning point for Pashto in the digital era. They promise easier education, services and communication for millions while preserving the language’s rich culture as artificial intelligence advances
The Qehwa achieved 85.3 per cent accuracy on a custom test covering 15 areas, including translation, culture, history and geography.
It understands prompts in Pashto, English or Urdu and replies naturally in Pashto, making it useful for education, content creation, government services and daily digital use.
“Pashto speakers deserve to interact with AI in their own language,” Mr Ahmad told this scribe. “I developed Qehwa to prove that even with limited resources, we can create something meaningful for our people.”
Fellow graduate Mohammad Uzair created Katib, ASR (automatic speech recognition), a strong speech recognition system for Pashto, and trained it on one of the largest Pashto voice datasets — with help from Mozilla Common Voice — it records a word error rate of 28.23 per cent, the best for open-source Pashto speech tools.
“The hardest part was not the model, but the data,” Uzair noted. “Pashto lacks clean recordings. I built systems from scratch to prepare them. Katib is the first step to give Pashto a voice in AI.”
Both projects were completed independently, without money, teams or official backing.
This shows the strength and creativity of Pakistan’s youth, and young people should draw inspiration from the duo, remarked Ali Jan, an IT expert.
Shafeeq Gigyani, a technology activist for social good, welcomed the initiative. “Pashto has been neglected not because of fewer speakers, but lack of resources,” he said. “Volunteers collected voice data through Mozilla Common Voice. It is heartening to see young innovators like Junaid and Uzair develop such AI tools. This is how we grow inclusive technology.”
Qehwa LLM and Katib ASR are open-source and freely available on Hugging Face for developers and researchers worldwide, the young developers said.
Published in Dawn, April 11th, 2026
