Leading an M.Tech thesis on Generative Retrieval Systems with a focus on PLAID dense retrieval optimization. The project advances hybrid retrieval pipelines by fusing generative models, sparse-dense fusion, and accelerated inference strategies, targeting sub-second responses for large-scale knowledge bases.
Current milestones include designing curriculum-aware contrastive sampling, optimizing GPU memory footprints for large transformer decoders, and evaluating retrieval quality across multilingual benchmarks. Collaborating with research peers to operationalize findings for IIT Bombay's digital learning infrastructure.
