Rakuten X DataScience SG

For Sep 2026, we are collaborating with Rakuten to showcase their research! Do note the entry instructions, and event space is held on 19 FLOOR! How do you shrink a multi-billion-parameter LLM down to something that runs on your phone without breaking it? For our September meetup, a Rakuten research scientist takes us through the full journey from large language models to on-device AI. Agenda 7:00 PM – 7:30 PM Registration & Networking 7:30 PM – 8:30 PM From Large Language Models to On-Device AI: Making Models Smaller Without Breaking Them 8:30 PM – 9:00 PM Q&A and Closing ## Synopsis What does it take to turn a large language model into one that can run efficiently on a phone or laptop? This talk explores the practical journey of taking models ranging from hundreds of millions to several billion parameters, making them smaller and faster through training, compression, and optimization, and eventually deploying them on real devices. But making a model smaller is only half the challenge. How do we know that the compressed model still behaves like the original—and that any changes we see are intentional rather than caused by problems in training or deployment? The talk will walk through the model development pipeline and practical techniques for validating model quality and detecting unexpected changes along the way. ## Speaker Bio Sunil Kumar Yadav is a Research Scientist in the Frontier Research Department at Rakuten, where he contributes across the full LLM development pipeline for Rakuten AI series of models - data, evaluation, and training - alongside work on the on-device deployment of appropriately sized models, where privacy and cost make it both feasible and desirable. Before the LLM era, he worked on neural machine translation at Rakuten. He holds a Master's degree in Cybernetics from the School of Systems Engineering, University of Reading.