Bengali.AI

AI & ML interests

Bengali.AI is a non-profit research organization working towards democratizing Bengali language research. Our team consists of researchers active in industry, academia and the government who are working towards releasing datasets and models in different domains of Bengali language research. Grad students and faculties from different universities lead many of the projects and mentor the young researchers here. We also do outreach activities. 2017-2018: Organizing Bengali.AI Computer Vision Challenge; the first Kaggle competition from Bangladesh. 100+ teams participated (In collaboration with Third Space lab, UToronto). 2019-2020: Organized a Featured Kaggle Competition on Bengali Handwritten Grapheme Recognition. Over 2k teams participated in the competition and fought for a prize money of 10k USD. (In collaboration with Google). 2021-2022: Currently we have seven ongoing research projects. Some of these are under review process and some are in the dataset construction phase. I am mentioning some here: a. Bengali Document Layout Analysis b. Grapheme based Indic OCR (7 languages). c. A Bengali Constituency Parsing Corpus (Largest Bengali constituency parsing corpus construction ongoing, using a new annotation platform, that we are also opensourcing. ) d. Bengali Spell and Grammar checker e. 450+ hours of Publicly available Bengali Speech corpus. In collaboration with Mozilla commonvoice initiative.