Bhubaneswar: While global technology giants race to build massive artificial intelligence (AI) systems, regional Indian languages are often left behind as an afterthought. Changing that narrative for Odisha, Bhubaneswar-based AI researcher and entrepreneur Arun Misra has developed OdiaSLM family — a suite of Small Language Models built from the ground up for the language.The landmark release includes three distinct models: two foundational engines (OdiaSLM-60M-Base and OdiaSLM-100M-Base) and a conversational AI assistant (OdiaSLM-100M-Instruct) capable of answering questions, summarising news and drafting content in fluent Odia.Unlike existing AI tools that simply translate English data or bolt Odia onto massive foreign models, Misra built the OdiaSLM series from scratch, training it on more than 1.5 billion words of pure Odia text. This work was published in journal Zenodo on Nov 12.“Most global AI models are designed primarily around the English alphabet. When they process a rich, classical script like Odia, they chop single words into tiny, awkward fragments. This makes them slow, expensive to run and prone to grammatical errors or ‘hallucinations’ — where the AI invents false information,” said Misra.To solve this, Misra engineered a custom Odia vocabulary system that understands how Odia words and matras naturally fit together. As a result, OdiaSLM reads and generates Odia text 3.6 times faster and far more efficiently than global multilingual models.Even more remarkably, these models do not require expensive supercomputers or an active internet connection. Weighing just over 200 megabytes, OdiaSLM-100M-Instruct runs smoothly on an everyday laptop CPU. Furthermore, the conversational model has been specially trained on Odisha’s unique heritage — meaning it can speak accurately about Konark Sun Temple, Lord Jagannath’s Rath Yatra, Chilika Lake and even the cultural significance of Pakhala, alongside science, health and general knowledge.“The practical benefits for the state are immense. Because the AI operates completely offline on basic hardware, rural schools can deploy it as a personal Odia tutor in classrooms without internet connectivity,” said Misra, adding farmers and citizen service centres can access instant information on agriculture and healthcare in their mother tongue, while local journalists and offices can summarise documents and keep sensitive data 100% private.For a classic language spoken by over 45 million people, Odia’s long-term vitality in the digital era depends on having its own sovereign AI infrastructure. “To accelerate this ecosystem, we have released all three models as open-source resources — free for academic research, students, educators and non-commercial initiatives,” Misra said.