Jiakun Fan
MLSys, efficient LLM inference, and RSI infrastructure.
I am a Computer Science Ph.D. student at Virginia Tech, advised by Dimitrios S. Nikolopoulos, with expected graduation in May 2028. I am currently a Research Scientist Intern at ByteDance in San Jose, working on efficient LLM inference.
My research connects model algorithms with the systems that make them useful: diffusion-drafted speculative decoding, memory-efficient LLM serving, heterogeneous CPU-GPU execution, and reliable long-horizon AI research loops. I am especially interested in agents that improve through feedback and the infrastructure needed to run, evaluate, and reproduce their experiments.
I am seeking research internships for Summer 2027 in these areas.
Download my CV · Google Scholar · LinkedIn · Email
Research
- Long-horizon autonomous research infrastructure: coordinated GPU experiments, candidate evaluation, experiment lineage, and recovery from execution failures.
- Efficient LLM inference: diffusion-based speculative decoding, memory-aware serving, and heterogeneous execution.
- Agentic LLM serving: high-level research on KV cache management for long-context, multi-turn workloads.
Selected publications and manuscripts
Factorization-Guided Training and Inference for Diffusion-Drafted Speculative Decoding
Jiakun Fan, Min Xiao, et al.
dLLM-Serve: Bridging the Memory Gap in Diffusion Language Model Serving
Jiakun Fan, Yanglin Zhang, Xiangchen Li, Dimitrios S. Nikolopoulos.
APEX: Asynchronous Parallel CPU-GPU Execution for Online LLM Inference on Constrained GPUs
Jiakun Fan, Yanglin Zhang, Xiangchen Li, Dimitrios S. Nikolopoulos.