From Sampled Outcomes to Capability Distributions: Rethinking Supervision for LLM Routing
EMNLP 2026, 2026
Curriculum Vitae
An undergraduate researcher in the Big Data Communication Experimental Program at Nanjing University, working across model routing and computational communication.
Education
- B.A. in Big Data Communication, Nanjing University, 2024–Present (Expected 2028)
- Big Data Communication Experimental Program, School of Journalism and Communication
- GPA: 4.44 / 5.0 · Rank: 2 / 16
Research Experience
- Model Routing and Evaluation, 2026–Present
- Research on supervision, preference-based evaluation, and budget-aware inference for LLM routing.
- Contributed to ORBIT, DARS, and RouteJudge through benchmark construction, system development, and evaluation.
- Computational Communication, 2026–Present
- LLM-based simulation of policy attitudes using large-scale synthetic personas.
- Study of behavioral adoption in intelligent payment infrastructure.
Publications
ORBIT: An Optimal Routing and Budgeted Inference Toolbox
Frontiers of Computer Science, 2026
RouteJudge: Preference-Based Evaluation of LLM Routers under Pluralistic User Preferences
ICML 2026 Workshop, 2026
Preprints
Priors in LLM Routing: From Initialization to Regularization
Preprint, 2026
Selected Practice
- NJU Xiao Lanjing — Content planning, photography, filming, editing, and post-production; Outstanding Member, 2025–2026.
- Global Field Study · Singapore & Malaysia — On-site visits, communication, and field research with the enterprise group.
Technical Skills
- Programming: Python, SQL, C++, JavaScript
- Data & AI: pandas, NumPy, scikit-learn, LLM inference, model routing, statistical modeling
- Media: Content planning, photography, filming, video editing, data visualization