Erwin Wu

[Tagline TBD: who you are in one line]

About

I'm a PhD student at the School of Design, The Hong Kong Polytechnic University. I study phygital interaction and human-AI interaction, and my dissertation asks how AI can generate floor plans that respect hard physical constraints and can be relied on. Before the PhD I built and tested software at Alibaba, Shopee and Huawei. These days I also turn ideas into tools people use, such as VideoScript and AI Developer Dashboard, and I share hands-on notes about AI and development on X.

Works

  • VideoScriptPlatform

    Paste a video or podcast link and get the full timecoded transcript.

    It works with YouTube, Bilibili, Apple Podcasts, Spotify and Xiaoyuzhou, even when a video has no captions. Copy the transcript straight into Claude, ChatGPT or your notes, or export it as Markdown, SRT, VTT or TXT. New accounts get 120 free minutes.

    Open site

  • BTC Bear-Bottom Evidence DashboardPlatform

    Sixteen on-chain indicators that show how deep a Bitcoin bear market runs and how far the bottoming has come.

    Instead of guessing the exact low, it treats a bear-market bottom as a process and reads it on two axes. The pressure axis shows how much stress the market is under, and the bottoming axis tracks whether evidence of capitulation, seller exhaustion and fresh demand is adding up. Thresholds come from a manual review of past bear markets, and the AI only interprets within those facts. It is a research tool, not trading advice.

    Open siteView on GitHub

  • AI Developer DashboardSoftware

    A local desktop dashboard that gathers project progress from Claude Code, Codex and Git in one place.

    When several projects are in flight, it reads the activity these tools leave on your machine and merges it into project cards, a stage board and a timeline, then picks the three things worth doing today. On demand, AI can summarise where each project stands, what comes next and what is blocking it. It runs on Windows and macOS.

    View on GitHub

  • docs-by-versionSkill

    An Agent Skill that sorts project docs by version, keeping plans apart from what actually shipped.

    Over time, project docs mix what was planned with what was done. It splits each version into a requirements file and a completion file, keeps progress and a version index in docs/README.md, and ships a script that checks the structure. It works with Claude Code, Codex, Cursor and other tools that read SKILL.md.

    View on GitHub

  • RMF-EDSoftware

    Real-time object detection in low light that fuses RGB, near-infrared and LiDAR.

    Annotated near-infrared data is scarce, so a NIR-GAN first generates near-infrared images from RGB datasets. The detections from RGB, near-infrared and LiDAR point clouds are then fused. On KITTI it matches or beats existing fusion methods and takes only 21 ms. This is the code that accompanies the paper.

    View on GitHub

Papers

  • RMF-ED: Real-Time Multimodal Fusion for Enhanced Target Detection in Low-Light Environments

    Yuhong Wu, Jinkai Cui, Kuoye Niu, Yanlong Lu, Lijun Cheng, Shengze Cai, Chao Xu

    IET Cyber-Systems and Robotics · 2025

    PDFDOI

  • ThermOuch: A Wearable Thermo-Haptic Device for Inducing Pain Sensation in Virtual Reality through Thermal Grill Illusion

    Haichen Gao, Shaoyu Cai, Yuhong Wu, Kening Zhu

    SIGGRAPH Asia 2024 Conference Papers · 2024

    PDFDOI

  • Enhancing Immersion in Virtual Spaces: PhytualBlend - a real-time interaction system for seamless physical feedback

    Tiantian Lo, Yuhong Wu, Gerhard Bruyns, Daniel K. Elkin

    eCAADe 2024 (Data-Driven Intelligence) · 2024

    PDFDOI

  • PACS: Phygital-Aided Construction System based on Modular Customizable Timber Fabrication

    Tian Tian Lo, Yuhong Wu, Terrence Michael Curry, Chaozan Wu, Zitian Lai, Yangyang Pan

    CAADFutures 2025 · 2025

    View on PolyU Research Portal

Articles

Experience

  1. 2024-09 – present

    School of Design, The Hong Kong Polytechnic University

    PhD student

    I study phygital interaction and human-AI interaction. My dissertation asks how AI can generate floor plans that respect hard physical constraints and can be relied on.

  2. [年份] – [年份]

    Shopee, Shenzhen

    Software development and testing intern

    [One line TBD]

  3. [年份] – [年份]

    University of Liverpool

    MRes, Pattern Recognition and Intelligent Systems

    [One line TBD]

  4. [年份] – [年份]

    Huawei Consumer BG, Dongguan

    Intern

    [One line TBD]

  5. [年份] – [年份]

    Harbin Institute of Technology

    BEng, Mechanical Design, Manufacture and Automation

    [One line TBD]

  6. [年份] – [年份]

    Alibaba, Guangzhou

    Game development and testing intern

    [One line TBD]

Contact