At the Internet, Data and Society Lab @ LUMS, I’m building a post-hoc confidence-estimation wrapper for a pretrained fundus vision–language model, producing calibrated uncertainty signals for diabetic retinopathy grading without retraining the model. I also designed a randomized controlled trial to test whether sharing those signals with graders can mitigate automation bias.
At AI Hub for Maternal, Newborn & Child Health @ LUMS, I own the ETL layer for a prospective mother–infant cohort and contribute computational modeling for early risk stratification. I’m also opening a research line testing whether obstetric ultrasound features predict maternal and neonatal complications.
My work spans computational social science and information retrieval. I led a project examining GenAI-powered microtargeted persuasion and built its multimodal content pipeline; I also co-authored a personalized information retrieval pipeline using BM25, ColBERT, and frequency–recency weighting.
