cv

B.S.

University of Illinois Urbana-Champaign (UIUC)

Computer Science and Mathematics

• Algorithms
• Systems Programming
• Networks
• Database Systems
• Data Structures
• Computer Architecture
• Probability
• Numerical Analysis
• Programming Language & Compilers
• ML
• NLP

Research Intern, MSRA, Data, Knowledge and Intelligence Group

Microsoft

• Research on LLM Agent/Multi-modal for Microsoft Copilot in Excel. • Built TextAnalysisSFT, a multi-threaded pipeline generating SFT training data for the new TextAnalysis API in Excel Copilot: Kaggle mining (2000+ real-world CSVs) → heavy-text filtering → query & Office.js code gen → SheetEngine validation—improving Office Script code-gen accuracy by 75%.

Research Assistant (Advisor: Prof. Jiaxuan You)

U Lab, University of Illinois Urbana-Champaign

• SeeingEye (First Author & Project Lead): Designed an agentic information flow utilizing a small VLM ReAct agent to translate VQA-tasks into structured intermediate representations (SIR) for text-only LLMs, enabling multimodal reasoning that beat larger end-to-end VLMs. Built SmartGridCaption—a visual chain-of-thought / iterative caption-and-patch module. • OpenManus-RL (Researcher): Implemented REACT reasoning architecture (think→act→tool-use) and multi-agent flow for action optimization. Built MCP-compatible tool framework. Constructed SFT datasets for AgentBench (Gaia, WebArena, WebShop, AlfWorld).

Software Engineer Intern, Wechat Group

Tencent

• Developed a cross-platform (Android & iOS) hardware profiler that reports performance data for Wechat Mini Program, supporting 200+ partner teams to surface performance bottlenecks. • Built a Unity Mono Memory Profiler that found 40+ hidden allocations per pilot app and cut leak-related crashes by 120%.

Software Engineer Intern

NetEase Game

• Developed a SQLite + Flask + Vue asset DB that indexes multi-terabyte game art assets in < 200 ms. • Implemented a multi-threaded, Redis-cached auto-scanner that accelerates folder-to-DB ingest by 300%. • Rendered interactive mesh-to-level dependency graphs; tech artists now pinpoint broken links in < 15 min instead of 4h.

Programmer (Advisor: Prof. Dan Cermak)

The stu/dio at UIUC

• Developed games in Unreal, Unity, and Godot, including VRchaeology, Resilient Community, and RPG. • Researched and implemented VR interaction mechanics and performance optimizations.

Research Assistant (Advisor: Prof. Misha Sra)

University of California, Santa Barbara

• Designed multi-agent social simulation framework to model social contract formation. • Implemented persona-encoding via prompt engineering and personalized QA sets for differentiated agent personalities. • Developed agent economic system and AI policy generation framework.

LLM Software Engineer Intern

Reborn Network

• Developed a multiplayer Role-playing agents Unity VR Game enabling text/voice/embodied VR actions (e.g., holding hands, dancing) with real-time voice interactions (< 1s latency), boosting average session length by 38%. • Introduced vector-DB long-term memory (dynamic CRUD + RAG) so agents remember players and past events, raising dialogue-coherence ratings from 2/5 to 4/5. • Created a shareable "character-card" framework, seeding a UGC ecosystem and trimming designer NPC-setup time by 300%.

Where LLM Agents Fail and How They Can Learn From Failures

arXiv preprint arXiv:2509.25370, under review at ACL 2026

Kunlun Zhu, Zijia Liu, Bingxuan Li, Muxin Tian, Yingxuan Yang, Jiaxun Zhang, Pengrui Han, Qipeng Xie, Fuyang Cui, Weijia Zhang, Xiaoteng Ma, Xiaodong Yu, Gowtham Ramesh, Jialian Wu, Zicheng Liu, Pan Lu, James Zou, Jiaxuan You.

Understanding and Mitigating the Bias Inheritance in LLM-based Data Augmentation on Downstream Tasks

arXiv preprint arXiv:2502.04419, under review at ACL ARR 2025

Miaomiao Li, Hao Chen, Yang Wang, Tingyuan Zhu, Weijia Zhang, Kaijie Zhu, Kam-Fai Wong, Jindong Wang.

On the Democratic Society of LLM Agents

PersonaNLP Poster, First Workshop on LLM Persona Modeling (PersonaNLP), NeurIPS 2025 Workshop. Mexico City.

Gordon Dai*, Weijia Zhang*, Devang Acharya*, Jinhan Li*, Qianyu Zhang, Victor (Qibin) Huang, Ryan Zhou, Keyi Wang, Dennis Niu, Misha Sra.

SeeingEye: Agentic Information Flow Unlocks Multimodal Reasoning in Text-only LLMs

arXiv preprint arXiv:2510.25092

Weijia Zhang*, Zijia Liu*, Haoru Li*, Haoqi Chen*, Jiaxuan You. Designed an agentic information flow that utilized a small VLM ReAct agent to translate VQA-tasks into structured intermediate representations (SIR) for text-only LLMs, enabling multimodal reasoning that beat larger end-to-end VLMs.

Artificial Leviathan: Exploring Social Evolution of LLM Agents Through the Lens of Hobbesian Social Contract Theory

The First Workshop on AI Behavioral Science, ACM SIGKDD 2024

Gordon Dai*, Weijia Zhang*, Jinhan Li, Siqi Yang, Srihas Rao, Arthur Caetano, Misha Sra.

Exec & Workshop Lead

UIUC ACM Gamebuilders

Led weekly workshops on game development topics including computer graphics, Blender, Unity, C#, game AI, and performance optimization. Provided one-on-one technical mentoring to members.

Course Assistant, CS 233: Computer Architecture

University of Illinois Urbana-Champaign

Held office hours and one-on-one help sessions on pipelining, caching, memory hierarchy, etc. Co-created homework puzzles and helped build and debug the Spimbot MIPS assembly competition.

RecSys 2025, Reviewer

RecSys

Dean's List

University of Illinois Urbana-Champaign

Global Game Jam 2023 — Ranking 9th

Global Game Jam

Ludum Dare 2023 — Top 10%

Ludum Dare

UIUC MechMania28 AI Hackathon — 2nd Place

UIUC ACM