Rui Zhu

My research focuses on building DNA-level AI tools that are both high-performance and privacy-preserving. On the efficiency front, I design customized DNA large language models (LLMs) for early prediction of aging-related diseases, and engineer model-compression and long-context strategies that enable accurate reasoning over very long genomic sequences while lowering computational cost and the barrier to use (e.g., distillation, quantization, sparse/linear attention, and memory mechanisms). In parallel, I develop safeguards across the AI supply chain to protect individuals’ genomic data—advancing privacy-preserving learning and inference (federated learning, differential privacy and multi-party computation), data provenance tracking, and policy-aware access controls. Together, these efforts aim to translate genomic signals into clinically actionable, early risk assessments at scale, without compromising personal security, privacy, or trust.