I am currently an undergraduate student in the School of Data Science and Engineering (DaSE) at East China Normal University (ECNU), majoring in Data Science and Big Data Technology (2023–2027).

My research interests focus on AI agents, reinforcement learning (RL), and multi-agent systems (MAS) and their real-world applications. I am fortunate to work as a research assistant in the PLANING Lab under the supervision of Prof. Xiang Li, where I explore agent-based reasoning and skill orchestration for complex problem solving.

I am currently working at ByteDance (China Transaction & Advertising · Traffic Monetization Engineering) as an AI Full-Stack Engineer, working on AIGC platform engineering, client-side resource scheduling, and agent-based prompt optimization. Previously, I built LiveCompose — an RL-powered on-device photography composition assistant published on the App Store.

🔥 News

2026.07: 🚀 Started as an AI Full-Stack Engineer at ByteDance Traffic Monetization Engineering, working on AIGC platform engineering, client-side resource scheduling, and Agent development.
2026.06: 🎉 AIAdStream project recognized as an Outstanding Project at ByteDance Engineering Training Camp.
2026.06: 📱 LiveCompose (构妙) iOS app officially launched on the App Store!
2025.12: 📑 LiveCompose project completed with a research paper, recognized as a National-Level Innovation Project.
2025.03: 🎯 Started LiveCompose — an RL-based on-device photography composition assistant (National Innovation Project).
2024 & 2025: 🎓 Received ECNU Excellent Student Award and University Scholarship for two consecutive years.

💼 Internship

ByteDance — China Transaction & Advertising · Traffic Monetization Engineering
AI Full-Stack Engineer | 2026.07 – Present | Shanghai, China
- AIGC Platform Engineering: Delivered full-stack improvements across platform configuration, access control, collaboration, and analytics, and built an automated AI search result-collection workflow; long-form capture completed within 30 seconds, while six core metric groups supported continuous quality tracking.
- Client Resource Scheduling Architecture: Built a unified scheduling layer for five categories of content preloading tasks, using loading latency, hit rate, and runtime stability as core metrics to reduce resource contention and performance fluctuations; online experiments delivered substantial gains.
- Agent Development: Developed an agent-based reverse prompt optimization workflow that distills high-quality examples into reusable domain skills and introduces generation, evaluation, and iterative refinement, reducing repeated manual prompt tuning.

🚀 Projects

LiveCompose (构妙): RL-Based On-Device Photography Composition Assistant
App Store • National Innovation Project • 2025.03 – 2026.06
Role: Project Lead & Sole Developer
Highlights: Designed an Actor-Critic RL architecture with a GAIC aesthetic scoring model as the reward signal. Trained and deployed the model to iOS via CoreML for real-time on-device inference (crop box IoU: 79.68%). Built a 17K-image professional composition dataset released on HuggingFace. Developed the full iOS app in Swift and published it to the App Store.
GitHub Hugging Face App Store
On-Device LLM for Personalized Copy Generation
ByteDance Internship Project • Model Technical Solution Owner • 2026.07 – 2026.08
Tech Stack: Python, PyTorch, Knowledge Distillation, SFT, INT4 Quantization
Highlights: Led the model technical solution for a personalized on-device LLM, designing a 68M-parameter Qwen3-based architecture with GQA, RoPE, SwiGLU, RMSNorm, and tied embeddings. Built teacher-data generation and quality-filtering pipelines; distillation and supervised fine-tuning improved task pass rate from 69% to 78.3% and reduced logical-error cases from 41 to 15. Quantized the model to approximately 37 MB and collaborated on the mobile inference lifecycle, meeting a 25-second end-to-end latency target for inputs up to 2,500 tokens.
AIAdStream: AI-Powered Smart Advertising Feed iOS Client
ByteDance Engineering Camp • Outstanding Project • 2026.05 – 2026.06
Tech Stack: SwiftUI + MVVM + SQLite
Highlights: Built a single-column ad feed with three card types (big image, small image, video) and strategy pattern architecture. Integrated conversational AI search with Function Calling (5 tools: search, detail, similar ads, web search, AI enhance). Added "Fun Rewrite" — AI-powered ad copy generation in 4 styles with typewriter animation. Designed a comprehensive analytics dashboard with KPI metrics, daily trends, and channel distribution. Achieved 38 XCTest unit tests, full Dark Mode support, and iOS 26 Liquid Glass design compliance.
GitHub

🏆 Awards & Honors

Outstanding Project Award — ByteDance Engineering Training Camp (AIAdStream), 2026
National Second Prize — 2nd OpenAtom Competition, Complex-Scenario Visual Analytics
ECNU Excellent Student Award & University Scholarship — Consecutive years 2024, 2025

🛠️ Skills

Programming: Python, C, Go, C++, Swift, Kotlin, PyTorch, JavaScript
Agent Development: Following current Agent technologies including Claude Code, DSH, and Harness

🎓 Education