The National Education Policy (NEP) 2020 recommended that children be taught in their mother tongue, local language or regional language at least until Grade 5, and preferably till Grade 8 and beyond. The logic behind the recommendation was simple. Children learn best in the language they speak at home.Experts who drafted the policy, led by former ISRO chairman K. Kasturirangan, argued that young children grasp concepts and non-trivial ideas more quickly in their home language than in an unfamiliar one. UNESCO too has repeatedly highlighted the issue, estimating that 40% of the global population still lacks access to education in a language they understand.A parent intervenesFor Chennai-based parent and former TCS employee Ganga Ponnu, the problem became apparent while watching her own children learn. “They remembered stories, not textbook definitions,” she said. But much of the digital science content available to them was either entirely in English or consisted of dubbed cartoons that felt distant from their everyday lives.So, she built her own. Using a single smartphone and a suite of AI tools, Ms. Ponnu created an educational channel centred on two animated siblings: Lokesh, a younger brother with endless questions, and Varsha, his elder sister who patiently explains the world around him.The pair speak not in formal textbook Tamil but in Tanglish (Tamil and English) commonly heard in urban Chennai households. Gravity is explained while eating murukku. Lessons on digital safety are woven into stories involving Krishna. The siblings draw inspiration from her children’s everyday experiences, including wearing glasses, while their conversations mirror those taking place around her dining table. “They are not translations of existing characters,” she said. “They belong here.”Ms. Ponnu’s experience reflects a broader shift underway in Indian education. Creating educational content no longer requires a studio, an animation team or a publishing house. Increasingly, it requires an idea and a set of AI tools. The project is built entirely by her on her smartphone. She uses AI tools for visuals, scripting and editing, moving between multiple models depending on the task at hand. In six months, her channel has more than 60 lessons.The most meaningful feedback, she said, has not come from view counts. A parent in Chennai’s Virugambakkam told her that her quiet child had begun speaking up in class after watching Varsha explain food waste in a Class 1 EVS lesson. A user from Michigan wrote online that the sibling dynamic reminded him of children he had known decades earlier.Until now, a child in Chennai learnt gravity through apples falling from trees rather than murukku slipping from a plate. Stories travelled across languages but rarely across cultures.Boon of democratisationAccording to education entrepreneur Viplav Baxi, founder of AmplifiU (pedagogy teacher platform) the importance of local language learning is already well established in educational research. Students learn better in the language they are comfortable with, he said, because otherwise they are forced to simultaneously understand both the concept and the language used to explain it. “The language itself becomes an obstacle to learning outcomes,” he said.Local language educational content existed long before generative AI arrived, Mr. Baxi notes. Teachers and creators across YouTube and social media have been creating content in regional languages for years. The difference now is democratisation. “You don’t have to be an expert at creating videos anymore to create high-quality content,” he said.The government, too, has sought to address India’s multilingual learning challenge through technology. Launched in 2017, DIKSHA (Digital Infrastructure for Knowledge Sharing), the Ministry of Education’s national digital infrastructure for school education, hosts textbooks, teacher training modules, assessments and other curriculum-aligned resources in multiple Indian languages. Several State governments use the platform to distribute digital learning material and support classroom teaching.Experimenting beyond limitationsA Bengaluru-based AI platform also uses generative AI to create multilingual educational tools for aspirants of competitive exams. The platform combines an AI tutor with AI-generated educational videos in regional languages for students from Class 8 onwards preparing for examinations such as JEE and UPSC. The company has signed MOUs (memoranda of understanding) with the Karnataka and Punjab governments, according to co-founder Apurv Mehra. “More than 10,000 students have used the platform so far”, he said.Yet, Mr. Mehra flagged that multilingual AI remains a work in progress. Models frequently make grammatical errors around gender in regional languages, while translations can sometimes lose their intended meaning or context. “It’s not just about translating words correctly. Sometimes the meaning itself gets lost in translation,” he said.AI is also changing how educational material itself is designed. At a teaching academy’ AI-powered learning platform, teachers can upload curriculum requirements and receive not just information on what to teach but suggestions on how to teach it. The system takes into account variables such as learning pace, demographics and city tiers while generating material in English, Hindi and Tamil. Twenty-six additional languages are currently being tested.The platform is currently being used by around 4,000 students across schools in Madhya Pradesh and Hyderabad. “The focus is pedagogy,” said K. Manojkumar, the company’s chief AI officer.Creating multilingual content, however, is only one part of the challenge. Making it sound natural is another. For Mukil Vannan, founder of educational video company, localisation presents challenges that go far beyond vocabulary. His company converts textbooks and course material into explainer videos across all 22 scheduled Indian languages and has worked with governments including Tamil Nadu and Odisha.Pronunciation remains one of the biggest hurdles. Sanskrit words frequently confused the model until the company trained it using voice samples from a teacher, eventually improving accuracy significantly.Even within languages, geography matters. “The Kannada spoken in Mysuru is different from the Kannada spoken in northern Karnataka,” Mr. Vannan said. Capturing those differences while ensuring that content remains widely understandable remains a challenge.Making AI-generated educational content feel natural extends beyond vocabulary and pronunciation. Anil Golecha, co-founder of a B.tech CSE platform, pointed to another limitation. Human speech contains pauses, filler words and imperfections that make conversations feel natural. AI-generated speech often lacks those qualities. The company has experimented with introducing small imperfections into its models after early testing with the founders’ own children suggested that perfectly polished speech often felt unnatural.