Pakistan’s first homegrown GPT model launched by woman-led tech firm

ISLAMABAD, Pakistan (MNTV) — Pakistani tech firm Data Vault has launched the country’s first indigenous generative pre-trained transformer (GPT) model, ‘Zahanat AI,’ marking a significant step in Pakistan’s artificial intelligence (AI) landscape. 

The initiative is led by Mehwish Salman Ali, co-founder and CEO of Data Vault, highlighting a rare female leadership role in the country’s tech industry.

Zahanat AI is a text-based generative AI model designed to facilitate human-like conversations, answer queries, and assist across multiple domains. Developed to cater to regional needs, it represents Pakistan’s push toward locally trained AI solutions.

The model has been in development for years and has been operating from a dedicated data center in Karachi since 2022. Unlike AI models that rely on external cloud infrastructure, Zahanat AI processes all its data within Pakistan, ensuring local data security and sovereignty. The facility is equipped with high-speed internet and robust Distributed Denial-of-Service (DDoS) protection to safeguard operations.

Woman-led AI innovation

Under Mehwish Salman Ali’s leadership, Zahanat AI has taken a unique approach to cost-efficient AI development. The model operates on a hybrid GPU architecture, utilizing Nvidia GPUs alongside other processors. Initially, the project relied on used gaming GPUs to build its computational power, making it a cost-effective alternative to international AI models.

While China’s DeepSeek AI reportedly required $5 million in investment, the exact cost of Zahanat AI’s development has not been disclosed, though its creators claim it was built at a fraction of that cost.

“The problems and solutions it will provide are tailored to Pakistan’s culture and environment—it is truly a local model,” said Syed Muhammad Yahya, Chief Technology Officer (CTO) of Data Vault, at the model’s launch event in Islamabad.

Zahanat AI is built on Meta’s Large Language Model Architecture (LLaMA) and features 1.5 billion parameters. The model follows an open-source framework, similar to China’s DeepSeek AI, allowing for greater customization and local adaptation.

The project was privately funded – through friends and family – underscoring the growing interest in homegrown AI solutions within Pakistan’s tech sector.

“It’s not just about features,” Mehwish Salman Ali emphasized. “It’s about representation — showing that we can build world-class technology from Pakistan, led by Pakistanis.”

One of the most anticipated developments is the upcoming Z2 model, which will introduce support for Urdu and multiple regional languages, significantly expanding accessibility for local communities.