This UAE-built AI model understands Arabic and is better than ChatGPT, claims creator

Khaleej Times

20-05-2025

As artificial intelligence continues to reshape industries worldwide, the development of locally built, culturally attuned technologies is gaining momentum.
In the UAE, one company is tackling a long-standing challenge: accurate Arabic speech recognition. CNTXT AI, based in Dubai, recently unveiled Munsit — a speech-to-text model designed specifically for Arabic, trained on thousands of hours of regional audio data.
We spoke with Mohammad Abu Sheikh, CEO of CNTXT AI, about why building this technology in the UAE matters, the complexities of Arabic dialects, and what this means for the future of AI in the region.
What inspired you to build an Arabic speech recognition model here in the UAE, when global giants already dominate the field?
We built Munsit because global tech giants weren't solving our problem. Arabic voice tech has long been underserved. Most models are designed for English, then retrofitted for Arabic, leading to low accuracy and misunderstood dialects. We saw a clear need and felt a responsibility to act. The UAE, with its AI-first vision and infrastructure, offered the ideal launchpad. It's a country committed not just to adopting AI, but building it.
That's what led to Munsit: a model engineered from the ground up for Arabic, reflecting our dialects, our data, and our region. We wanted to accelerate the shift from AI consumer to AI producer.
Many talk about the limitations of Arabic in tech, but few have tackled it at this scale. What were the linguistic or cultural challenges you faced, and how did you overcome them?
While many see Arabic as too complex for AI, we see it as a strategic opportunity. The real challenge wasn't the language, it was the data. Less than 5 per cent of online content is in Arabic, and even less is usable for training. If data is the new oil, then unstructured data is oil unrefined — full of potential but useless until processed. Without high-quality data, you can't build high-performing models, so we solved this problem ourselves.
We developed a data pipeline from scratch using weak supervision — a scalable, algorithmic approach that processed over 30,000 hours of raw Arabic audio and refined it into a clean, high-quality dataset ready for large-scale training. That gave us the foundation to train Munsit on how Arabic is actually spoken, at a speed and cost traditional methods simply couldn't match.
How did you source such a large and representative Arabic dataset, and what did you learn about the voices of the region in the process?
We built our Arabic speech dataset from scratch, sourcing voices from a wide range of real-world environments — news broadcasts, casual conversations, public archives, and everyday interactions across the region. We captured dialectal variation and quickly realised we were documenting the lived experience behind the language. These differences, shaped by history, geography, and culture, are more than linguistic. They're expressions of identity and belonging.
CNTXT AI calls this a 'sovereign technology' — what does that mean for the UAE's place in global AI development?
Sovereign AI means full ownership of the data, the infrastructure, and the outcomes. In the UAE, that translates into national investment and AI readiness at every level. Munsit is a result of that vision: built locally, deployed securely, and aligned with the country's digital priorities.
The UAE is defining its own path in AI; building models that reflect regional identity and serve local needs. Data sovereignty is central to that mission. Data is precious, and it must remain in our hands. That's how the UAE moves from participant to standard-setter in global AI — exporting trusted, culturally grounded technology.
What does this breakthrough mean for everyday Arabic speakers, especially in education, public services, or content creation?
Arabic speakers now have a model that understands them in real time, with contextual accuracy and speed. In education, it enables dialect-aware tools for early learners and non-literate users. Imagine Emirati ed-tech platforms offering voice feedback that reflects how students actually speak. In government, it addresses dialect diversity, especially in judicial settings where interpretation can break down. Munsit detects these differences, transcribes accurately, and localizes output into formats like Emirati Arabic. It powers fast, scalable transcription and indexing in media, making Arabic content easier to find, distribute, and monetise.
How big of a role did homegrown talent play in building Munsit, and do you see this as a turning point for young AI developers in the UAE?
Munsit was shaped by homegrown talent — every layer reflects regional hands and regional voices. And yes, this is a turning point. You don't need to leave the region to build breakthrough AI. The infrastructure is here. The capital is here. The ambition is here. The ecosystem is ready. You can invent, and not just implement, from the region and lead globally. It's validation for the next generation: world-class AI can, and will, be built right here.
What comes next for Munsit and for Arabic voice AI as a whole?
What's next? A new generation of Arabic-first products, designed here and deployed globally. Munsit serves as the voice layer in our broader AI stack alongside tools for preparing, testing, and deploying AI in a sovereign way. From this foundation, we're expanding fast: domain-specific voice agents and multilingual dialect switching. One of the most exciting developments : our Arabic Text-to-Speech suite, launching with Emirati and Saudi dialects. With native voice talent onboarded, we're delivering the region's fastest, most accurate Arabic TTS, a major step toward full-stack voice infrastructure.
What would you tell a young developer or linguist in the UAE who dreams of building world-class tech, right here?
Start now. Move fast. You don't need permission. You're already in one of the most AI-ready nations on earth. So build. Don't just dream of catching up. Dream of leading. Because if we don't build the future in our language, solving our own problems, who will?