Jixian Su's Homepage

Jixian Su's Homepage

she/her

Ph.D. Student @ Georgia Tech jixiansu(at)gatech(dot)edu

I am currently a PhD student in Computer Science at the Georgia Institute of Technology, advised by Prof. Kexin Rong. Previously, I earned my B.Eng. in Computer Science and Technology from Shanghai Jiao Tong University (SJTU) in 2024, where I had the privilege of working with Prof. Shixuan Sun. I am broadly interested in the intersection of databases and machine learning, particularly data systems for understanding, evaluating, and improving LLM prompts and pipelines.

Publications

  • Prompt Analytics [paper] [project website]
    Jixian Su, Yusen Zhang, Eugene Wu, Kexin Rong.
    Under submission.
    We introduce prompt analytics, a research paradigm for identifying reusable prompt strategies and analyzing where they are used, when they transfer, and how they affect model behavior and task performance. Our system, FACET, transforms unstructured prompt collections into structured, queryable data linking prompt strategies to their usage contexts and observed effects.

  • Revisiting the Design of In-Memory Dynamic Graph Storage [paper] [code]
    Jixian Su*, Chiyu Hao*, Shixuan Sun, Hao Zhang, Sen Gao, Jiaxin Jiang, Yao Chen, Chenyi Zhang, Bingsheng He, Minyi Guo.
    Proceedings of the ACM on Management of Data (SIGMOD’25).

  • RapidStore: An Efficient Dynamic Graph Storage System for Concurrent Queries [paper] [code]
    Chiyu Hao, Jixian Su, Shixuan Sun, Hao Zhang, Sen Gao, Jianwen Zhao, Chenyi Zhang, Jieru Zhao, Chen Chen, Minyi Guo.
    Proceedings of the VLDB Endowment (VLDB’25).

*: Equal contribution.

Education

  • Georgia Institute of Technology
    Ph.D. in Computer Science, Aug. 2025 – Present.
  • Shanghai Jiao Tong University
    B.Eng. in Computer Science and Technology (Zhiyuan Honor Program of Engineering), Sept. 2020 – June 2024.
    GPA: 3.86/4.3, Rank: 10/106.

Research Experience

  • Prompt Analytics for LLM Systems (FACET), Georgia Tech
    Research Assistant with Prof. Kexin Rong, Aug. 2025 – Present.
    • Introduced prompt analytics, a research paradigm for identifying reusable prompt strategies and analyzing where they are used, when they transfer, and how they affect model behavior and task performance.
    • Architected and implemented FACET, an end-to-end system that transforms unstructured prompt collections into structured, queryable data linking prompt strategies to their usage contexts and observed effects.
    • Built a prompt corpus and reusable feature library of 35,188 prompts from 22,298 authors across 31 public sources, spanning research code, prompt hubs, custom GPTs, workflow templates, and agent skills, and identified 24,288 features that recur across authors.
    • Measured the impact of individual prompt strategies through 6.5M controlled comparisons that added or removed one strategy while holding the rest of the prompt fixed. Reused strategies with positive evidence to improve 10 of 11 Text-to-SQL settings by up to 4.6 points and warm-start automatic prompt optimization, improving held-out Text-to-Cypher accuracy by 0.7–10.3 points.

  • Dynamic Graph Storage Systems, Shanghai Jiao Tong University
    Research Assistant with Prof. Shixuan Sun, Nov. 2023 – June 2025.
    Architected a composable C++ testbed that decouples graph containers, neighbor indexes, and concurrency-control protocols; reimplemented four index designs and two protocols for controlled, component-level evaluation. Diagnosed memory amplification, cache inefficiency, and lock contention through white-box experiments on five dynamic graph storage methods and eight graphs.

  • Survey and Benchmarking of DGS Methods, National University of Singapore
    Research Assistant with Prof. Bingsheng He and Prof. Shixuan Sun, Mar. 2023 – Nov. 2023.
    Developed a storage-design taxonomy and unified benchmarking framework for comparing performance, memory efficiency, scalability, and concurrency trade-offs across in-memory dynamic graph systems.