Mohammad Abu Sheikh is reworking the AI panorama within the MENA area, driving a shift from passive consumption to sovereign innovation. As CEO of CNTXT AI and founding father of a $10 million AI fund, he has led three profitable exits and secured over a billion {dollars} in funding. His work is laying the inspiration for an AI ecosystem rooted in language, tradition, and knowledge sovereignty.
CNTXT AI is a digital transformation firm that gives cloud infrastructure, industrial software program, and robotics options to assist organizations modernize operations and unlock data-driven insights throughout the Center East and North Africa.
What impressed you to begin CNTXT AI, and the way did your imaginative and prescient for sovereign AI within the Arabic-speaking world start?
We noticed the abundance of underutilized knowledge on this a part of the world. A number of issues in scaling AI got here from the dearth of knowledge readiness — which finally meant a scarcity of AI readiness. That’s why we began CNTXT AI.
Initially, we have been fixing the identical issues we confronted whereas constructing LocAI…We noticed these challenges firsthand working with AI71, TII and G42 (IIAI). As we helped these entities remedy these issues, the imaginative and prescient obtained clearer and the enterprise simply stored rising.
You’ve performed a key position in constructing the biggest Arabic digital library for AI coaching. What have been among the largest challenges in doing so, and the way did you overcome them?
High quality was one of many largest challenges. One other was the restricted availability of high-quality Arabic knowledge on-line: Arabic is severely underrepresented. Solely a small portion of Arabic-language content material has been digitized, and simply 3–5% of all on-line content material is in Arabic. That’s virtually nothing. We overcame that downside by deploying knowledge labelers, annotators, and knowledge scientists to digitize, create, and curate the info ourselves.
CNTXT AI operates on the intersection of tradition and computation. How do you steadiness cutting-edge AI innovation with the objective of constructing culturally related options for the MENA area?
We construct culturally grounded fashions from the bottom up. From infrastructure to closing product, tradition is embedded from the very starting — it’s not one thing we add later. We design, innovate, and construct with particular cultures, dialects, and wishes in thoughts from day one. Arabic is one language, but it surely carries many dialects and cultural contexts throughout the area, so we construct native merchandise for native nations. And we do this by working with native annotators, folks on the bottom, in their very own nations.
You’ve got additionally co-founded LocAI and lead the SMPL AI Fund. How do these ventures complement the mission of CNTXT AI?
LocAI is the applying layer — the half folks truly work together with. It sits proper on high of the info and infrastructure constructed by CNTXT AI. That’s what made it profitable: it transforms AI foundations supplied by CNTXT AI into real-world options folks can use.
SMPL AI, then again, is about giving again to the group. It focuses on investing in early-stage startups and serving to construct the regional AI ecosystem. We share the instruments and classes we’ve discovered from constructing AI ourselves, so founders can develop sooner and keep away from frequent pitfalls.
Munsit has been referred to as probably the most correct Arabic speech recognition mannequin on this planet. What drove the event of this mannequin, and why now?
What drove the event of this mannequin was easy: the necessity.
We at all times construct out of necessity. We regarded on the market and noticed the panorama was ripe — authorities businesses and personal purchasers have been all asking for an answer like this.
The prevailing fashions simply weren’t as much as the duty. Most are constructed on English tech after which tailored. They aren’t designed for Arabic from the bottom up, and positively not for the precise issues we’re fixing.
So we determined to construct our personal. It’s Arabic first — by design.
The analysis behind Munsit introduces a weakly supervised studying strategy. Are you able to clarify what meaning and why it was important for coaching Arabic ASR at scale?
Annotation is pricey. So we needed to transfer past conventional strategies that rely upon giant quantities of guide transcription. Weakly supervised studying helped us scale with out having to label each audio file by hand — which is particularly necessary for Arabic, a language with restricted knowledge and many various dialects.
As an alternative of utilizing professionally transcribed audio, we began with 30,000 hours of unlabeled Arabic speech. We constructed an annotation pipeline that generates, filters and cleans the most effective ones utilizing automated checks. This gave us a high-quality 15,000-hour dataset — all with out human transcription.
This strategy made it potential to coach our mannequin from scratch, capturing the richness of spoken Arabic throughout real-life conditions, rapidly and cost-effectively. With out this technique, constructing an Arabic ASR system at this scale would have taken years and thousands and thousands in guide effort.
Munsit outperformed fashions from OpenAI, Microsoft, and Meta throughout a number of benchmarks. What does this achievement say about the way forward for Arabic AI innovation?
The way forward for Arabic AI is in our palms; and that’s precisely what this achievement proves. We are able to not afford to depend on applied sciences we don’t personal or rely upon third events who don’t prioritize our area.
Munsit exhibits that we are able to construct world-class AI, from the area, for the area — utilizing native expertise to unravel native issues. It’s a transparent sign that the subsequent wave of Arabic AI innovation will come from inside.
How do you see Munsit evolving in future variations, and what are the subsequent frontiers for Arabic voice AI at CNTXT?
You’ll simply have to attend and see. What I can say is that now we have a contemporary, new suite of Arabic-first AI options on the way in which — all powered by Munsit and different fashions we’re presently constructing at CNTXT AI. That is only the start.
You usually converse in regards to the significance of “sovereign AI.” What does that time period imply to you, and why is it crucial for the Gulf and broader MENA area?
To me, sovereign AI means having full possession and management over the info, infrastructure, and fashions that form our future. It’s crucial as a result of we have to personal our personal destiny, and that begins with knowledge.
Knowledge sovereignty is all the things. Knowledge is valuable, and we want to verify it stays in our palms.
We are able to’t afford handy over our future and sit idle whereas others construct the know-how for us. The way forward for AI on this area will come from this area. That’s precisely what we’re working towards.
How do you see CNTXT AI shaping the AI ecosystem within the Center East over the subsequent 5 years?
By enabling true AI readiness. We go in, perceive what corporations and governments want, construct the info and AI methods, after which assist them construct, check, deploy and scale.
If knowledge is the brand new oil, then unstructured knowledge is oil unrefined — filled with potential however ineffective till processed. That’s why we’ve constructed CNTXT AI to assist organizations clear, construction, and activate their knowledge. As a result of that’s the place actual AI transformation begins.
Out of your vantage level as each an entrepreneur and investor, what recommendation would you give to different founders constructing AI startups in rising markets?
Begin now. Transfer rapidly. Fail quick, study sooner, and hold iterating.
Most significantly, construct for actual issues. Keep near the bottom — hearken to customers, not simply the hype. In rising markets, relevance and adaptableness are key.
Thanks for the nice interview, readers who want to study extra ought to go to CNTXT AI.
