My research centers on personalized generation & retrieval and large language models (LLMs) for code. I design personalized ranking models that address data sparsity, representation collapse, and cross-domain shift in recommender systems, and I build LLM agents that migrate and translate code at repository scale while aligning multi-modal LLMs with real-world tail signals. More recently, I explore proactive, just-in-time agents that infer when to act and re-rank content on-device, without an explicit request.