My research interests are in human-computer interaction (HCI) and information, communication technologies and development (ICTD) and Data for social good. In my work, I frequently use qualitative and quantitative methods. My goal is to contribute to HCI research by designing, building, deploying, and evaluating computing technologies and to help improve the lives of underserved communities.
(c) Data-driven prediction, luck and hunch in betting:
Decolinizing AI in the Global South
Gender, Harassment, and Transformative Justice
I examine the opportunities and issues that arise in designing technologies to support low-income rural women in Bangladesh. I conduct qualitative studies to explore the systemic everyday challenges women face that form the backdrop against which technology design could potentially happen. In this ongoing research, I investigate women's harassment on Social Media and design technologies to support gender justice.
(Mis)Information, Rural Computing, and Social Wellbeing
I study wellbeing practices and misinformation among the rural communities in Bangladesh to design more usable and culturally appropriate technologies.
Faith, Stigma, and Access to Computation
I explore the social and cultural factors that influence the online betting practices among the villagers and how bets harmonize with users’ faith, hunch, and cultural practices, along with statistical recommendations. Shada Bakso is a hardware device, designed by us, to explore the fears of using mobile phones among the rural women of Bangladesh. My probe study with the rural women using Shada Bakso suggests that rural women's fear of technology further initiates technology non-use.
Speech Emotion Recognition(SER)
Previously I worked on SER. I concluded that to design a speaker-independent SER system should focus on speaker-independent features of speech signals. My frequency-based analysis provided significantly higher accuracy than many of the state-of-art works. My algorithm produced 80.55% correct results while the best known previous algorithm had 56.98% of success.