Past Projects
  • A Large-scale Entity Resolution

    • The broader goal was to maximize quality in real world data by identifying duplicates. My primary role was to research, design and develop a highly scalable entity resolution system. The system achieved near the state-of-art accuracy (DBLP-ACM: 0.973), low latency (~17 mins on 6M records) and high scalability and was featured by Towards Data Science.

  • Mercedes Benz: MBUX Voice Control

    • The broader goal was to assist drivers with voice control. My primary role was to lead the development of NLU system as the Point of Contact (PoC) for the NLU Research team. I worked closely with other PoCs (ASR, LM, Dialog and TTS) to keep track of end-to-end KPIS such as accuracy, quality, coverage and latency for 20+ language NLU development. My job was to ensure meeting SLAs before delivering product update.

  • BMW: Voice Command

    • The broader goal was to assist drivers with voice control. My primary role was to lead the development of NLU system as the Point of Contact (PoC) for the NLU Research team. I worked closely with other PoCs (ASR, LM, Dialog and TTS) to keep track of end-to-end KPIS such as accuracy, quality, coverage and latency for 20+ language NLU development. My job was to ensure meeting SLAs before delivering product update.

  • Audi: Voice Recognition

    • The broader goal was to assist drivers with voice control. My primary role was to lead the development of NLU system as the Point of Contact (PoC) for the NLU Research team. I worked closely with other PoCs (ASR, LM, Dialog and TTS) to keep track of end-to-end KPIS such as accuracy, quality, coverage and latency for 20+ language NLU development. My job was to ensure meeting SLAs before delivering product update.

  • Tesla: Voice Commands

    • The broader goal was to assist drivers with voice commands. My primary role was to lead the development of NLP, Dialogue and NLG systems in English and Korean languages for driving. Also, I developed the phonetic matching system to allow entity resolution (i.e. songs, contacts and places, etc.) across different languages.

    Voice Commands is now used in 1.3 million Tesla vehicles.

  • LG Electronics: Voice Mate

    • The broader goal was to assist smartphone users with an intelligent personal assistant in English and Korean languages. My primary role was to directly engage with LGE from product demo, product roadmap, product development to product delivery for their new smartphone release, LG G4.


  • BlackBerry: BlackBerry Assistant

    • The broader goal was to assist smartphone users executing tasks using natural voice. My primary role was to lead the develpment of NLP system in English supporting over 20 domains and 100 intents for virtual assistant application. Our team managed to successfully deploy on BlackBerry's new flagship device, Passport.

  • QNX: HTML5 WebPlatform

    • The broader goal was to support web developers to create a mobile application using HTML5 SDK or Apache Cordova. My role was to build the common ground, WebPlatform, for all web applications and browser to ensure that applications and UIs are renderred accordingly per device event (i.e. Touch, Gesture, Rotation, ContextMenu, etc.) across smartphones, tablets and automotives.

    Today QNX OS is now used in 215 million vehicles.

  • Combined Forces Command: Ulchi Freedom Guardian

    • Ulchi Freedom Guardian is the world's largest computerized command and control implementation, involving 50,000 South Korean troops alongside 17,500 U.S. troops and mainly focuses on defending South Korea. My primary role was to work on ground intelligence.

    Obtained top-level security clearance.