• Tech leader for Llama4 speech model evaluation, including infra, benchmarks, metrics to assess the model performance from various aspects including understanding, generation and conversational dynamics for both pre-training, post-training and RL

• Led and executed evaluation pipelines for speech tokenizers and audio encoders, measuring both semantic accuracy and paralinguistic representation.

• Tech leader for pronunciation modeling, pronunciation learning, personalization and contextual biasing of ASR models, shipped to various interfaces including Ray-Ban Meta glasses and voice-in-feed

• Delivered the system of real-time ASR decoding based on Kaldi

• Built deep learning models to detect stress and evaluate vowel and consonant quality

• Led the group for automatic pronunciation assessment including stress, vowel and

consonant for L2 English speech

• Architected and designed new generation DMA, NAND controller and NAND PHY

• Developed Firmware for command fetch of Host for NVMe PCIE3

• Responsible for Firmware development of NAND controller, working closely with

NAND engineer and achieving the required performance for mass production

• Built the first generation NAND model of our company using C++ from scratch

• Brought up the first generation verification environment of NAND controller using C++

• Designed and architected the main modules of data path including SOVA controller, scrambler, RLL and LDPC encoder