Weiyi Wang is a staff software engineer at Google developing LiteRT, the framework that runs machine-learning and generative-AI models on phones, computers, and embedded devices. His work makes on-device AI more practical across incompatible chips and operating systems, with particular attention to inference speed, battery consumption, and hardware portability.
In 2025, Wang contributed to LiteRT’s ahead-of-time compilation tooling and was among the foundational contributors to Google AI Edge Gallery, an open-source application for running models locally, including offline speech transcription. That November, he coauthored Google’s introduction of the LiteRT Qualcomm AI Engine Direct accelerator, which gives developers a common interface for accessing specialized mobile hardware and compiling models either in advance or on-device.
His technical priorities include:
- Neural processing unit acceleration: Making dedicated AI chips accessible through consistent deployment workflows, while accounting for the battery life, thermals, responsiveness, and device variability that determine whether applications work outside controlled benchmarks.
- On-device function calling: Contributing to Google AI Edge Gallery’s local tool-use capabilities, enabling compact models to trigger application actions while retaining the privacy, responsiveness, and offline availability of local inference.
- Cross-platform edge deployment: Supporting AI applications across Android, iOS, and embedded hardware. At AI Engineer Europe 2026, Wang joined Chintan Parikh for audience questions on target devices and execution approaches; Parikh delivered the principal presentation.
Wang has also led a Google I/O Connect China 2026 workshop combining Google AI Studio, MediaPipe, and LiteRT to help developers build and deploy on-device applications.
