The following paper has been accepted at Practical ML for Low Resource Settings Workshop (PM4LRS) @ ICLR 2024.