Bangla AI: Scarce Data, Abundant Promise
Spoken by over 230 million people, Bangla remains under-served in AI research.
Rafiq Ahmed6 min read
Illustration — The Bengal Diplomat
Large language models are trained mostly on English data, so answers in Bangla can be uneven, especially for regional dialects and technical subjects.
Local start-ups and universities are building open datasets. The hard part is collecting literature, news and spoken recordings while respecting ethics and copyright.
If it works, the impact is large: from farm advice to healthcare, services in the mother tongue.
Sample story — written for demonstration.