Uyir Lab newsroom

TORONTO, October 8, 2026. Uyir Lab, an independent AI lab, today introduced Language Brain, a layer that makes AI agents accurate in the languages large AI models get wrong, the way those languages are actually spoken.

Large language models learn mostly from the written web, while most people speak in dialects, slang and mixed languages that rarely appear in training data. As businesses put AI agents in front of customers for banking, health, telecom and public services, those agents misunderstand hundreds of millions of people. Language Brain sits between an existing agent and the people it serves, interpreting dialect, slang and code-switching and helping the agent reply naturally.

Uyir Lab is starting with Tamil, spoken by more than 86 million people. Its first model, Nila, is trained from scratch on 9.6 billion tokens of Tamil, English and Tanglish. Tamil speakers worldwide contribute voice recordings through the Uyir contribution studio, where every recording draws one loop of a kolam, the traditional Tamil floor drawing. Contributions are made with explicit consent and reviewed by people before use.

After Tamil, Uyir Lab plans to add Sinhala, Punjabi, Swahili, Hausa, Amharic, Javanese, Fula, Yoruba, Bhojpuri, Oromo, Pashto, Igbo, Khmer, Haitian Creole and Quechua: 16 languages with more than 854 million speakers. The lab reuses its pipelines and evaluation methods across languages, so each new language costs less than the last.

Businesses building AI agents for these markets can request early access at uyirlab.com. Tamil speakers can contribute at uyirlab.com/contribute.