My final-year thesis at MIST (Department of CSE) has the title "Analysing Low-Resource Language Feedback Across Industries". It uses several machine learning approaches and a custom dataset to detect sentiment and emotion in Romanized Bengali.
What Romanized Bengali is, and why it's hard
Romanized Bengali is Bengali in Latin letters, the way people write when they text, post and review things. It has no standard spelling, so the same word shows up in several forms. Most sentiment tools target English, or Bengali in its own script. Few labelled datasets exist for it, which is what "low-resource" means here.
Why feedback
The title points at the use: feedback from customers across industries. A model that can read it can tell a business what people like and what upsets them, in the language they actually write in.
Where it led
I finished the thesis in 2025. These days my M.Sc. asks the next question: adaptive control of local LLM agents on consumer hardware.