Hello, I'm

Frank Zeng

ML/AI Engineer & Tech Lead

I build scalable AI infrastructure and ML systems — from algorithm design to global production deployment.

About Me

Frank Zeng

I'm a seasoned ML/AI engineer with 15+ years of industry experience spanning generative AI infrastructure, ads quality systems, computer vision, and large-scale ML deployment. Currently leading video generation serving infrastructure at ByteDance.

Former Google Senior Software Engineer with deep expertise in ML-powered ads traffic quality. Strong foundation in research and production systems — from algorithm design to scalable, low-latency global deployment. Ph.D. from Tsinghua University.

0+ Years Exp.
0+ ML Filters Launched
0+ $M Fraud Prevented

Skills & Tech

ML & AI

LLMsRL (PPO, GRPO)SFT RLHFLoRA/QLoRAAgentic AI

Training & Inference

TransformerMoEDeepSpeed vLLMTensorRTRay

Infrastructure

KubernetesDockerAWS CUDAPostgreSQLRedis

Computer Vision

CNNObject DetectionObject Tracking OpenCVSEM Image Processing

Experience & Education

Employment

Nov 2024 – Present ByteDance San Jose, CA

Tech Lead Software Engineer, GenAI

  • Led development of scalable video generation serving infrastructure supporting global production deployment of state-of-the-art generative models.
  • Designed and optimized distributed pipelines delivering low-latency, high-throughput inference for cutting-edge video generation models.
  • Applied model optimization techniques to reduce serving costs while maintaining generation quality at scale.
GenAIVideo GenerationDistributed SystemsModel Optimization
Jul 2017 – Nov 2024 Google Mountain View, CA

Senior Software Engineer

  • Led the development and launch of 10+ sophisticated machine learning filters (clustering, classification, self-supervised learning) to detect and prevent over $30M ARR in ad fraud traffic, safeguarding ecosystem health.
  • Strategically initiated and led innovative ML initiatives with director-level support, overseeing 10 engineers in deploying graph-embedding and BERT-based techniques for semantic analysis and entity detection.
  • Directed cross-functional collaboration with interdisciplinary teams of 10+ researchers and analysts, managing the end-to-end ML lifecycle from prototype development to global deployment and monitoring.
  • Mentored over 7 engineers and interns, fostering a culture of technical excellence and leading deep-dive sessions on domain knowledge and AI applications within the team.
MLBERTGraph EmbeddingsAds QualityLeadership
Mar 2014 – Jul 2017 Hermes-Microvision, Inc. San Francisco, CA

Sr. Algorithm Software Engineer

  • Developed efficient algorithms for SEM image denoising, super-resolution, edge detection, and alignment for e-beam wafer/mask inspection applications.
  • Optimized algorithm performance for real-time, high-precision semiconductor inspection systems.
Computer VisionC++Image ProcessingReal-time
Jun 2012 – Mar 2014 Canon Information Technology (Beijing) Beijing, China

Image Processing Researcher

  • Developed a state-of-the-art real-time industrial object detection algorithm (face, head-shoulder) optimized for SOC implementation under strict accuracy, speed, and memory constraints.
  • Built a real-time HD camera surveillance system integrating object detection and tracking techniques.
Object DetectionObject TrackingC++Embedded

Education

2007 – 2012 Tsinghua University

Ph.D., Computer Vision & Image Processing

2003 – 2007 Huazhong University of Science and Technology

B.Eng., Communication Engineering

Experience Highlights

Video Generation Infrastructure

Led scalable video generation serving infrastructure at ByteDance, supporting global production deployment of state-of-the-art generative models with low-latency, high-throughput inference.

GenAIDistributed SystemsModel OptimizationKubernetes

Ad Fraud Detection at Google

Developed 10+ ML filters using clustering, classification, and self-supervised learning to detect and prevent over $30M ARR in ad fraud traffic.

MLBERTGraph EmbeddingsPython

Semiconductor Inspection Algorithms

Developed efficient algorithms for SEM image denoising, super-resolution, edge detection, and alignment for e-beam wafer/mask inspection at Hermes-Microvision.

Computer VisionC++Image ProcessingReal-time

Real-time Object Detection System

Built state-of-the-art real-time industrial object detection (face, head-shoulder) optimized for SOC implementation, plus HD camera surveillance system at Canon.

Object DetectionObject TrackingC++Embedded

Get in Touch

I'm always open to new opportunities and interesting projects. Let's build something great together!

frankzengai@gmail.com LinkedIn GitHub