For most people, the arrival of ChatGPT in early 2023 was an introduction to a new kind of technology. For Arushad Ahmed, it exposed a problem that had been quietly sitting underneath the AI revolution: what happens when the language you speak is not one of the languages the technology understands well?
That question would lead Ahmed, co-founder and CTO of Maldivian software company Javaabu, to build DhivehiGPT, an AI platform designed specifically around Dhivehi, the language spoken in the Maldives. What started as a weekend experiment in his dorm room at the University of St Andrews has grown into a broader language-technology platform spanning conversational AI, speech recognition, text-to-speech and optical character recognition. The scale of adoption is beginning to tell the story. DhivehiGPT has now passed 120,000 registered users, processed more than 3 million prompts, transcribed nearly 2,000 hours of audio and processed more than 200,000 OCR pages, according to Javaabu’s latest figures.
The bigger ambition, however, is not simply to build a Maldivian version of ChatGPT. It is to make sure Dhivehi has a place in the AI era.
From Microsoft Paint to Building a National AI Platform
Ahmed’s path into technology began long before artificial intelligence became fashionable. Around the age of six, after moving to Malé for school and encountering a computer at his cousin’s house, he became fascinated with computers through games and experiments with Microsoft Paint. Later, studying Computer Studies at Majeediyya School introduced him to programming and reinforced his natural inclination toward problem-solving.
That interest eventually became a career. Around 2010, while studying at the Centre for Higher Secondary Education, Ahmed and his friend Mohamed Jailam volunteered to build the school’s website. A few years later, their technical collaboration became a business when Jailam began bringing website projects to Ahmed, eventually leading to their first paid project for a guesthouse in Hulhumalé. Ahmed began freelancing, built a growing client base and continued developing his skills while studying Computer Science at the University of Nottingham Malaysia.
In 2015, the two friends won a Maldives Red Crescent website tender. When Ahmed returned to the Maldives after graduating in 2016, he briefly joined Dhiraagu, the country’s largest telecommunications company. He lasted roughly two weeks before deciding that building his own company was more compelling than a conventional career, particularly after Javaabu won a much larger information-system project for the Housing Ministry. By 2017, Javaabu had become a formal 50-50 partnership, with Jailam as CEO and Ahmed as CTO.
The company would go on to work on major national digital projects, including the Visit Maldives website, Job Center, eTukuri and the Kiyeveni e-learning platform. Ahmed has subsequently contributed to more than 100 websites, information systems and mobile applications, according to Javaabu.
Why Dhivehi Needed More Than a Translation Layer
Ahmed’s interest in AI was already developing before the current generative AI boom. His undergraduate final-year project involved optical character recognition, while Javaabu had undertaken early AI work including a text-to-speech project. He eventually pursued an MSc in Artificial Intelligence at the University of St Andrews, graduating with distinction and receiving the School of Computer Science Award for the highest overall performance.
His Master’s research focused on Dhivehi automatic speech recognition, forcing him to think about the language not simply as text but as a technological system requiring datasets, models, evaluation and specialised engineering. The timing was almost perfect: Ahmed arrived at St Andrews in September 2022, and ChatGPT appeared a few months later. By early 2023, he was experimenting with what would become DhivehiGPT from his dorm room.
The first version was deliberately simple. Ahmed connected Google Translate to a language model, translating Dhivehi into English, sending the English prompt to the model and then translating the response back into Dhivehi. It worked surprisingly well, so he built a basic interface and released it publicly for free. The experiment quickly became an overnight local hit.
But real users exposed the limitations almost immediately.
Translation could change the meaning of questions, particularly around local island names and geographic terminology. More importantly, an English-oriented global model could produce answers that were technically plausible but poorly aligned with Maldivian cultural, social and Islamic contexts. Ahmed realised that DhivehiGPT could not remain a translation wrapper around someone else’s model; if Dhivehi AI was going to become genuinely useful, the underlying infrastructure had to improve.
The Hardest Part of AI Is Sometimes the Data
Building AI for a language spoken by a relatively small population creates a problem that developers working on English or other high-resource languages rarely encounter at the same scale: the infrastructure has to be created before the AI can be improved.
Ahmed describes data scarcity as DhivehiGPT’s biggest technical challenge. For speech recognition, the team spent approximately six months and worked with multiple paid annotators to produce roughly 10 hours of carefully annotated audio. Achieving human-level performance could require datasets measured in more than 1,000 hours, making traditional data collection prohibitively expensive for a bootstrapped company.
The challenge extends beyond datasets. Dhivehi’s Thaana script has its own characteristics, requiring specialised text normalization, preprocessing and language-specific tooling. OCR can require synthetic data, while the team must also develop evaluation methods and benchmarks that are already well established for major languages.
This is why DhivehiGPT’s significance extends beyond the chatbot interface. Javaabu describes the platform as an ecosystem combining Dhivehi-focused language models with in-house speech-to-text, text-to-speech and OCR technologies. The company is effectively building pieces of a digital language stack that did not previously exist at the same level.
When AI Becomes Part of the Workday
The strongest evidence that DhivehiGPT has moved beyond novelty is how people use it. Ahmed says users increasingly turn to the platform for practical tasks such as drafting letters, translating documents, preparing written material and handling everyday office work. Usage patterns even correlate with the Maldivian working week, suggesting that the platform has become embedded in people’s productivity routines rather than remaining something users open simply to experiment with AI.
The numbers reinforce that shift. Javaabu recently reported that DhivehiGPT had reached approximately 120,000 registered users, three million prompts, nearly 2,000 hours of speech transcription and more than 200,000 processed OCR pages.
Its technology is also expanding. The latest text-to-speech engine includes seven new voices and supports Dhivehi, English and Arabic, giving the underlying technology applications beyond a single-language chatbot. Ahmed led the technical development of the upgraded TTS engine, working alongside Javaabu’s machine-learning and creative teams.
For Ahmed, however, the primary mission remains Dhivehi. The objective is to build the strongest possible AI experience for the Maldivian language while allowing technologies developed along the way to inform Javaabu’s broader international ambitions.
A Bigger Question for the Maldives
Ahmed’s work has also placed him in a position that extends beyond building products. He has contributed to national AI curriculum development and policy discussions, while Javaabu has participated in AI literacy initiatives for government and other institutions. In August 2026, Ahmed received a National Award of Recognition for contributions to communications and information technology, specifically in computer software and software application development.
His view of what the Maldives needs next is unusually practical. The country needs more investment in AI research, stronger university pathways, scholarships for AI and machine-learning fields, better datasets and benchmarks, and more opportunities for local startups to participate in government procurement. It also needs international payment infrastructure that allows Maldivian technology companies to sell software globally without payments becoming a major operational barrier.
For a country as small as the Maldives, competing through scale is difficult. Ahmed believes the answer is to compete through talent, innovation and specialised expertise.
That philosophy may ultimately prove more important than DhivehiGPT itself.
From DhivehiGPT to a Global Product Company
Ahmed is candid about the challenges of building a technology company in a small market. Javaabu has experienced cash-flow pressure caused by delayed government payments, difficulty retaining experienced engineers and the limitations of a domestic market with finite demand. During difficult periods, Ahmed and Jailam sometimes had to personally inject money into the business to ensure salaries were paid.
Yet DhivehiGPT has changed the trajectory of the company. What began as a project has demonstrated that Javaabu can build and operate products used at national scale, giving the founders confidence to move beyond a primarily project-based software business toward a product-focused company.
That transition is already visible in the company’s latest work. Javaabu recently launched DhivehiVaahaka.com, a platform for Dhivehi readers and authors that uses DhivehiGPT’s text-to-speech technology while giving local writers an opportunity to publish and earn from their work. Ahmed describes the broader ambition as building a Maldivian creator economy around Dhivehi literature.
For DhivehiGPT itself, Ahmed wants the platform to become unquestionably the strongest Dhivehi AI product and, eventually, a case study for what can be achieved in other low-resource languages. For Javaabu, he envisions a portfolio of technology products capable of competing internationally, alongside a stronger research culture that produces internationally published AI research.
There is an appealing symmetry to Ahmed’s story. He started by discovering what a computer could do as a six-year-old in Malé, learned to build software because a school needed a website, and eventually left a stable corporate job because he wanted to create technology of his own. More than a decade later, he is working on a problem that is simultaneously technical, commercial and cultural: ensuring that one of the world’s smaller languages is not left watching the AI revolution from the outside.
For Arushad Ahmed, building AI for Dhivehi is not about making the Maldives look like Silicon Valley. It is about making sure the Maldives has a place in the future being built there.
