VERSYS

Wilson Zhao is a staff-level ML engineer, creative technologist, and music producer based in New York. Primarily interested in developing "humanity-affirming" technology for live shows, his work, such as DDSP-VST (link), aims to elevate live experiences by intertwining tech with human emotion.

His artist name, Versys, stands for "versatile system," which reflects his artistic style and also happens to be the brand of motorcycle he crashed three years ago. Previously, he worked on continous learning/multimodal foundational models at Waymo and posttraining voice agents at Hume. Now, he's thinking **very hard** on how to model multimodal aesthetics at MSL.

VERSYS

赵申申是一位常驻纽约的高级机器学习工程师、创意技术专家和音乐制作人。 他主要致力于开发"人性化"的现场表演技术，他的作品，如DDSP-VST（链接）， 旨在通过将技术与人类情感交织来提升现场体验。

他的艺名Versys代表"多功能系统"，这反映了他的艺术风格，也恰好是他三年前撞毁的摩托车品牌。此前，他在Waymo从事持续学习/多模态基础模型的研究，以及在Hume从事语音代理的后训练。现在，他在MSL**绞尽脑汁**思考如何对多模态美学进行建模。

CREATIVE

  1. DDSP VST (2021-2022). Real-time AI timbre transfer.
    1. Description: Built the first real-time AI audio VST, which culminated in a performance in Google I/O 2022.
    2. Open-source repo [link]
    3. Google I/O performance (2022) [link]
    4. Silicon, Robert Laidlow (2022) [link]
    5. Technical deep dive at ADCx (2023) [link]
  2. Brelly (2023). Modular portable rave lighting system.
    1. Leave No Trace (2022-2023). 4-part ephemeral party series.
  3. Soundcloud (2023-now). WIP.

PROFESSIONAL

  1. CMU SCS Class of 2018
  2. Google X (2019) - Remote sensing using LSTMs to predict crop yield. [Patent]
  3. Chronicle (2019-2020) - Petascale security telemetry search and indexing
  4. Waymo (2020-2024) - Perception, O(100B) problem cluster discovery using contrastive embeddings, active learning, concept discovery, two patent applications, foundational models.
    1. Waymo One now open to everyone in San Francisco
    2. Rare Example Mining for Autonomous Vehicles
  5. Hume (2024-2025) - Posttraining speech-text transformers for emotive voice agents, with a focus on naturalness, aesthetic preference, and audio quality
    1. Website | EVI 2 Launch Video, 2024
    2. #1 on Huggingface TTS Arena (updated 5/12)
  6. Meta Superintelligence Labs (2025-now) - Multimodal (text, audio, video) real-time aesthetics

WRITINGS

  1. thesis_0.html

FAVORITE MULTIMODAL ARTIFACTS

  1. Game of Life Cellular Automaton (Wolfram) [semantic] [visual]
  2. Poly 2 (Ostgut Ton) [audio]
  3. Symphony in Acid [visual] [audio]
  4. @ybdg_4 (Instagram) [visual] [embodiment]
  5. Chichu Art Museum [embodiment] [visual]
  6. Pescadero [embodiment]
  7. OCaml Multicore Parallelism [semantic]
  8. DeepSeek-V2 Math [semantic]
  9. Barker X Baumecker (live) - Halle am Berghain [audio] [visual]

创意作品

  1. DDSP VST (2021-2022). 实时AI音色转换。
    1. 简介：开发了首个实时AI音频VST插件，并在Google I/O 2022上进行了表演。
    2. 开源代码库 [链接]
    3. Google I/O 表演 (2022) [链接]
    4. Silicon, Robert Laidlow (2022) [链接]
    5. ADCx技术深度解析 (2023) [链接]
  2. Brelly (2023). 模块化便携式派对灯光系统。
    1. Leave No Trace (2022-2023). 临时派对系列。
  3. Soundcloud (2023-至今). 进行中。

专业经历

  1. 卡内基梅隆大学计算机科学学院 2018届
  2. Google X (2019) - 使用LSTM进行农作物产量遥感 - 专利
  3. Chronicle (2019-2020) - 千万亿级安全遥测搜索和索引
  4. Waymo (2020-2024) - 感知系统，使用对比嵌入、主动学习、概念发现进行O(100B)规模问题集群发现，两项专利申请，基础模型。
    1. Waymo One现已向旧金山所有人开放
    2. 自动驾驶车辆的罕见样本挖掘
  5. Hume (2024-至今) - 用于情感语音代理的语音文本转换器后训练，专注于自然度、美学偏好和音频质量
    1. 网站 | 演示视频, 2024
    2. Huggingface TTS Arena排名第一 (更新于5/12)

写作

  1. thesis_0.html

最爱的多模态作品

  1. 生命游戏元胞自动机 (Wolfram) [semantic] [visual]
  2. Poly 2 (Ostgut Ton) [audio]
  3. Symphony in Acid [visual] [audio]
  4. @ybdg_4 (Instagram) [visual] [embodiment]
  5. 地中美术馆 (Chichu Art Museum) [embodiment] [visual]
  6. Pescadero [embodiment]
  7. OCaml 多核并行 (Multicore Parallelism) [semantic]
  8. DeepSeek-V2 Math [semantic]
  9. Barker X Baumecker (live) - Halle am Berghain [audio] [visual]
Wilson Zhao