For Mahad Omar, co-founder and CMO of Ilmiya, the value of AI is not “just about speed or scale,” he says. “It’s about the impact on people.” For the Ilmiya team, that impact means democratizing access to high-quality education for children around the world.

Ilmiya, an AI-powered education platform founded in 2021, is led by Taha and a team of 15 employees. The company has leveraged AI to deliver personalized education to thousands of students across five continents. The platform helps educators build, manage, and grow language learning programs by creating course content and providing student engagement and study tools. But as the platform grew, so did its challenges. The demands of creating course content for a rapidly growing student base, while also effectively assisting and motivating each student, began to exceed the team’s capacity.

The challenge: Balancing growth and student needs

The first challenge was generating enriching course content at scale. “Manually creating lesson plans, quizzes, and objectives for multiple subjects and education levels was slow and inconsistent in quality,” says Mahad. With a small team and thousands of students, tailoring guidance to each learner proved difficult. “Providing individualized study advice and motivation to students based on their performance data wasn’t feasible manually,” he explains.

Helping students stay on track was another key challenge. “Students struggled with time management and identifying which tasks to focus on,” notes Mahad, “especially when falling behind.” The rapid pace of course creation also introduced quality assurance issues. “We had to identify outdated, broken, or ineffective content, and that required manual review, slowing down updates.” The Ilmiya team knew they needed a better technical infrastructure to automate and enhance key learning workflows as they grew.

The solution: Smarter, automated workflows with the Gemini API

“We began by identifying pain points in our platform that could be addressed with generative AI,” says Mahad. A cross-functional group from their product, development, and academic teams identified automated course curriculum generation, personalized student coaching, motivational workflows, and content review as the most high-impact use cases.

The team first experimented with OpenAI’s GPT models but “encountered limitations with long context handling and cost-versus-performance trade-offs,” Mahad says.

They then integrated the Gemini API into their platform. “Our product team would identify feature requirements and define how Gemini model outputs would fit into user experiences,” explains Mahad. “The dev team would handle API integrations and prompt engineering.” Meanwhile, the academic team created content frameworks and reviewed AI-generated outputs, “helping tune prompts for pedagogical accuracy.”

Using course metadata—such as subject, grade level, topic, and regional curriculum—as input, “we were able to generate complete course module syllabi, lesson content, quizzes, and learning objectives with the help of the Gemini API,” says Mahad. The team primarily used Gemini 1.5 Experimental Pro for course generation. “We used this model when we needed more context-aware output,” he adds. “This allowed us to handle deeper academic logic and nuanced content generation.”

To better coach students, the team fed quiz performance and time-on-task data into the Gemini API to generate personalized study tips and motivational nudges. “This replaced the need for a human mentor for each student and led to improved retention,” Mahad says. To keep students on track, the team developed a status system they called Red-Amber-Green (RAG), which uses Gemini models to classify and prioritize quiz tasks as urgent, pending, or on track. The system can also develop short- and long-term study plans to help each student manage their own learning.

To automate content review, the team input incomplete lessons and broken quiz items into the Gemini API, which would then identify outdated or low-quality modules and suggest edits. For these real-time tasks, Mahad notes, “we used Gemini 2.0 Flash because of its low latency and cost-efficiency, which is critical for fast responses in a production environment with many concurrent students.”

The results: Faster development and higher retention

After integrating the Gemini API, the team saw development time drop and student learning outcomes improve significantly.

“We saw 70% faster course development cycles, and updates that previously took weeks now completed in days,” says Mahad, “plus a 25% increase in student retention and engagement.” Quiz performance improved by 10%, and lesson completion rates increased by 15% across pilot groups.

The Ilmiya team also achieved a 40% reduction in manual QA costs due to automated content reviews. “This allows our team to focus on innovation and student experience,” says Mahad. One instructor told him, “Gemini cut my review time in half. Now I spend less time fixing and more time coaching students.” These gains have helped the Ilmiya team focus on growing its business, with plans to expand globally “thanks to the volume and quality of content our team can create with the help of Gemini models,” Mahad adds.

What’s next: New ways of learning to help every student

The Ilmiya team is now planning to launch a next-generation AI tutoring assistant with a focus on ESL and reading fluency, in addition to enhancing its multimodal offerings. “Students could upload PDFs, notes, or images, and the Gemini models can create study plans and quizzes for them,” says Mahad.

The most important takeaway from the experience? “Think of Gemini models as infrastructure, not a feature. It gives startups enterprise-grade scale, long-context reasoning, and real-time personalization from day one.” When it comes to implementation, Mahad advises other founders to “start small with one high-impact workflow, track metrics early, and keep a human-in-the-loop for pedagogy until a system is proven reliable.”

Mahad and the Ilmiya team remain passionate about the potential for AI-powered education. “With tools like Gemini models, education is finally getting the upgrade it desperately needed.,” he concludes. “Every child, every teacher, and every parent, no matter where they are in the world, can have equal footing and access to opportunity. This is about equity.”