Bryan's Resumé

About

个人简介

From 🇨🇦 Toronto, Ontario 来自 🇨🇦 安大略多伦多

Lives in 🇺🇸 Los Angeles, California 现居 🇺🇸 加利福尼亚洛杉矶

Third-year electrical engineering major at UCLA. Previously engaged in product R&D at Unity and NLP research at Tsinghua University. Looking forward to connecting with amazing people and contributing to impactful projects in hardware technology, software development, robotics, and AI.

加州大学洛杉矶分校电气工程专业大三。曾在Unity从事产品研发，以及在清华大学从事自然语言处理研究。期待与优秀人才交流并为硬件技术、软件开发、机器人和人工智能领域有影响力的项目与研究做出贡献。

Experience

工作经历

Unity Technologies

Unity科技

Software Engineer Intern Internship

软件工程实习生实习

Shanghai, CHN · On-site

中国上海 · 现场

Jun 2025 - Aug 2025

2025年6月 - 2025年8月

Built major components of the PC core SDK of WebGL minigame hosting platform on Electron, encompassing, input logic, adaptive design, cross-platform support, and IPC
构建基于Electron的WebGL小游戏宿主平台电脑端核心软件开发工具包的主要组件，涵盖输入逻辑、自适应设计、跨平台支持及进程间通信
Drove key updates to its Android runtime SDK and sample activity on V8, enabling dynamic data fetching, optimizing caching strategies, and streamlining package loading to reduce TTID
主导其基于V8的安卓端运行时软件开发工具包及其示例应用的关键更新，实现动态数据获取，优化缓存策略并改进包加载以减小首帧绘制时间
Facilitated integration of its app with Unity Hub, establishing delegated game handle listening, and implementing parallel decryption and deferred download for WASM resources
推动其应用程序与Unity Hub的集成，建立游戏句柄委托监机制听并实现WASM资源的并行解密与延迟下载

Institute for AI Industry Research, Tsinghua University

清华大学智能产业研究院

Research Intern Internship

科研实习生实习

Beijing, CHN · Hybrid

中国北京 · 混合

Aug 2024 - Jan 2025

2024年8月 - 2025年1月

Conducted research on long-context processing strategies for LLMs to address context degradation, focusing on deep knowledge extraction and retrieval efficiency from 200K to 1M token literary corpora
开展有关大语言模型长文本处理策略的研究以解决上下文退化问题，聚焦于从20万至100万词元的文学语料库中进行深度知识提取及其检索效率
Designed benchmark suites spanning 60+ literary classics for rigorous evaluation of sliding-window, parameter-reduced, and RAG approaches on factual understanding and higher-order inference tasks
设计涵盖60余部文学经典的基准测试套件，用于系统评估滑动窗口、参数压缩及检索增强生成方法的技术路线在事实理解与高阶推理任务中的表现
Contributed to a literature Q&A agent for clients on an optimized GraphRAG framework, improving multi-hop reasoning accuracy by 23% over baseline models on complex unstructured narratives
参与为客户打造的基于创新的图检索增强框架的文学名著问答智能体的开发，在针对复杂非结构化叙事文本的多跳推理上其准确率较基线模型提升23%