Community Spotlight
Dr Pauldy Otermans & Dev Aditya on building offline AI teachers
By Ioanna Lykiardopoulou
Every month, we spotlight one member for our community. At KickstartAI, we believe that meaningful progress starts with individuals – builders, researchers, founders, and practitioners – who are applying AI in bold, practical ways. Our Community Spotlight series celebrates their work, their challenges, and the lessons they’re learning along the way.
This month, we’re featuring Dr Pauldy Otermans, co-founder and Principal & Head of AI Research at Otermans Institute, and Dev Aditya, co-founder and Managing Director/CEO & AI Lead. This is their story.
Dr Pauldy Otermans and Dev Aditya met at university in London. Sharing a passion for education, they traveled to South Asia to teach students in marginalized communities – and experienced first-hand the lack of sufficient educational support.
Their early work quickly led to the first government MOU from the state of Uttarakhand in India to teach and train students soft and employability skills. Soon after, the duo founded Otermans Institute with the mission to make learning accessible to all and upskill 750 million learners by 2030.
From fieldwork to scaling
When the pandemic hit two months later, Pauldy and Dev moved their learning programs online, teaching via Zoom with the support of teachers, trainers, and volunteers. “At some point in 2020, we reached 35,000 learners,” Dev says. “Since our focus was on actual teaching rather than simply providing content, we realized we needed a different approach to be able to scale.”
The team decided to experiment with AI around the same time that the first Transformer architecture papers were emerging, hinting at the new possibilities of the technology. In January 2021, they launched what they describe as the world’s first AI teacher – 18 months before the launch of ChatGPT.
“For the first lesson we created, we didn’t have enough training data,” Dev recalls. “So we wrote about 12,500 lines of questions and answers as a data layer.” Their first AI teacher, called Beatrice, was first used at BCF UNHCR refugee camps in Northern Iraq.
Fast-forward to today, Otermans Institute has expanded into 19 countries, reaching learners across the entire education spectrum – from primary and secondary school to higher education, skills development, vocational training, and national workforce programmes. “We’re content-agnostic,” Pauldy notes. “Our systems can teach almost anything.”
Some of their most significant recent work includes running Africa’s largest AI literacy programme, and a new agreement with the government of Burundi, where their AI teachers will support one million students from elementary school through high school.
Behind the Otermans Institute's AI teachers
Dubbed OIAI, the company’s conversational AI teachers are built on a combination of proprietary large and small language models. They follow Bloom’s taxonomy of learning, guiding learners through progressively deeper levels of understanding.
The system supports personalized learning and adapts to individual learning needs, including special education needs. “For those with learning difficulties such as dyslexia and ADD, autism, we’ve built in layers of accessibility,” Dev explains. This includes not only adjustments to the language the AI uses, but also a series of neurodiversity-friendly features – from background color and font size to subtitle styling, avatar type, and speaking speed.
While all AI teachers share the same architecture, their appearance and language changes depending on the country they’ll be used. “It could be the way they look, the accessories they have, and also the way they speak,” Pauldy explains.
But AI is only one part of the company’s technology stack, which also incorporates infrastructure, software as well as the end-consumer and data analytics layer.
In practice, the system begins with a dashboard where institutions and governments upload their curricula and course materials. The platform converts this content into structured lessons, which educators can then adjust or refine. Once approved, these lessons can reach learners through the AI teacher app.
“When we first started our average learner completion rate was 64%, which rose to 95% as of January 2025, including our free courses,” Dev adds.
Going offline
This year, the team made its biggest breakthrough yet: taking its AI teachers offline. “For marginalized communities the real bottleneck is connectivity,” Dev says. “Even when internet access exists, it often comes at a cost people can’t afford.”
By early autumn, the team had also reduced the offline app size from 1.2GB to 500MB.
In this setup, the phone effectively becomes the server, with the AI running directly on-device. That’s enabled by a mix of techniques: switching between LLMs and SLMs depending on the use case, and compressing the models through quantization.
But the decisive element, Dev says, is engineering. Rather than sending an entire user interaction to the model at once, the system breaks it into small, sequential steps – making the workload manageable even for a lightweight model. And behind this sits a runtime that switches components on and off as needed, ensuring the device’s limited memory is used efficiently.
“Our architecture is hybrid,” he explains. “90% of activity runs offline on the edge, directly on the user’s device, while 10% runs on the cloud.”
Ethics in AI teaching
This hybrid architecture plays a key role in the company’s ethical framework, the two founders explain. Because the system runs offline, learners’ personal data stays on their device rather than being sent to remote servers, reducing risks typically associated with cloud-based AI.
At the same time, the small part of the system that runs on the cloud and accesses the internet collects anonymous, aggregated data, which it feeds back to partner organizations. These analytics help institutions identify learning gaps and assess the effectiveness of their materials, informing decisions about educational content and policy.
This is particularly important for countries like Burundi, Uganda, and rural Kenya,” Dev says, “where we’re helping them capture learning data that was previously recorded only on paper.”
“Our aim is to make education accessible to everyone,” Pauldy concludes. “Our AI teachers are not here to take anyone’s job.They can be the first point of learning for people who would otherwise have no access to education, and they can support learners outside classroom hours when a human teacher simply isn’t available.”
