Machine Learning Research Lead

Machine Learning Research Lead — Apex AI Labs

San Francisco, CA | January 2021 – Present

• Directed a team of 8 AI research scientists to pre-train a 70B parameter large language model on 1.2T tokens, improving model downstream benchmark accuracy by 14% and reducing compute costs by $250,000 using pipeline parallelism.
• Designed and optimized a custom distributed training framework using JAX and Ray across 512 GPU nodes, which reduced training wall-clock time by 35% and increased training hardware floating point utilization to 62%.
• Implemented a low-latency model inference pipeline using Triton Inference Server and TensorRT, decreasing latency for real-time generative models by 45% and saving 28% in GPU hosting costs under high query volumes.
• Authored 5 papers in premier machine learning conferences including NeurIPS and ICML, securing 3 patents for novel transformer attention mechanisms that decrease key-value cache memory footprints by 30% during inference.

Senior Machine Learning Researcher — Vertex Research

Palo Alto, CA | June 2015 – December 2020

• Led the development of a production computer vision model for object detection, improving average precision by 18% and reducing inference latency from 95ms to 40ms on resource-constrained Edge platforms.
• Developed a multi-task learning pipeline for image and text representation, boosting classification accuracy by 12% across 6 downstream application products and reducing data preprocessing times by 34%.
• Pioneered neural architecture search techniques to automate model discovery, which accelerated research cycles by 38% and reduced manual hyperparameter tuning efforts from 14 days to less than 24 hours.
• Collaborated with 4 cross-functional product squads to integrate advanced reinforcement learning algorithms into the core recommendation system, boosting user daily active engagement time by 15% over 9 months.

Machine Learning Research, Deep Learning, Generative AI, Large Language Models (LLMs), Reinforcement Learning, Distributed Training, PyTorch, JAX, Ray, Triton Inference Server, CUDA, TensorRT, Kubernetes, Docker, Python, C++, SQL, Git

70B Parameter LLM Pre-training

Role: Research Lead

Tools: PyTorch, Megatron-LM, Kubernetes, AWS

Directed pre-training of a frontier 70B parameter model, improving benchmark scores by 14% and saving $250,000 in cluster costs.

JAX Distributed Training Orchestrator

Role: Core Researcher

Tools: JAX, Ray, GCP, TPU v4

Built custom parallelization framework, accelerating model convergence by 35% and supporting distributed training across 512 nodes.

• Google Cloud Certified Professional Machine Learning Engineer (2021)
• AWS Certified Machine Learning – Specialty (2020)

• Languages: English (Native), Mandarin (Conversational)
• Publications: 8 peer-reviewed papers in NeurIPS, ICML, and ICLR (2018-present)
• Availability: 4 weeks notice