Currently

Studying
M.S. Computer Science, Fudan University2025–2028
Research
LLMs for CodeAI for Software EngineeringCoding Agents
Before
B.Eng. Software Engineering, Tianjin University2021–2025
Based in
Shanghai, China

Contact

GitHub
alanzeng423
LinkedIn
alanzeng423
X
@azzz423

[email protected]

I study how coding agents can be trained, evaluated, and trusted on real software projects.

Alan Zeng

I’m a second-year M.S. student in Computer Science at Fudan University, working on LLMs for code and AI for software engineering.

I’m interested in how coding agents can be trained, evaluated, and trusted on real software projects. That means going beyond passing tests: respecting a project’s design constraints, and learning from the design knowledge buried in code review history.

Outside research I build small developer tools and AI-powered utilities for my own workflow. Most of them are on GitHub, and I write up what I learn, mostly in Chinese, at notes.alanzeng.com.

Research

  1. arXiv 2610.21925Oct 2026

    Benchmarking and Enhancing Design-Aware Code Review in the Linux Kernel

    DesignLore and DesignReviewBench reuse the design decisions recorded in past review discussions to guide an agent’s code review.

    relative F1 gain over the strongest baseline
    21.6%

    arXivPDFCopy BibTeX

  2. arXiv 2604.05955Apr 2026

    Does Pass Rate Tell the Whole Story? Evaluating Design Constraint Compliance in LLM-based Issue Resolution

    SWE-Shield, a benchmark for whether agent patches respect a project’s design constraints. Test pass rates substantially overestimate patch quality.

    issues
    495
    validated design constraints
    1,787
    repositories
    6

    arXivPDFCopy BibTeX

Side projects

  • VocabPocketA macOS menu-bar app for instant translation and OCR, with a vocabulary list and spaced review.Swift
  • Zotero PaperLensA Zotero plugin that reads the whole paper first, so a translated selection comes with the paper’s context.JavaScript

Experience

For a detailed CV, email me.

  1. bytedance/seed-coding.yaml

    role: LLM Algorithm Intern
focus:
  - Agent safety review
  - Browser-agent post-training
  - Long-context SFT
  - Async agents
  - Evaluation

    Shanghai, China

    ByteDancePDI-Seed-Coding

  2. sensetime/research.yaml

    role: LLM Algorithm Intern
focus:
  - SWE data scaling
  - SWE SFT
  - RL infrastructure
  - Test-case rewards

    Shanghai, China

    SenseTimeResearch Institute

  3. tiktok/monetization.yaml

    role: Frontend Engineering Intern
focus:
  - Ads platform frontend
  - Analytics
  - A/B testing

    Shanghai, China

    ByteDanceTikTok Global Monetization

Write to me about coding agents, evaluation, or code review. My full CV is available by email.

I read every message.

GitHubLinkedInX