Lead Machine Learning Engineer
Summary
Lead the development of foundation models for Grab’s marketplace, designing transformer architectures and scalable training pipelines using PyTorch and DeepSpeed.
Lead Machine Learning Engineer at Grab.
About the role
This position involves spearheading applied research and engineering efforts to develop foundation models for the Grab marketplace.
Key facts
Location: Singapore, One-North
Engagement: Full-time
Team: AI Automation Team
What you’ll do
- Create and manage scalable pre-training pipelines for multimodal and large language models, covering data curation to final checkpoints.
- Build and operate distributed training frameworks on multi-node GPU clusters using advanced parallelism and memory-efficient techniques.
- Develop generative recommendation systems that integrate pre-training with post-training loops such as SFT, distillation, and reinforcement learning.
- Design custom transformer architectures, including Mixture-of-Experts and long-context attention models, to fit specific business needs.
- Oversee code and design quality while mentoring junior staff to ensure production-grade standards.
- Collaborate with infrastructure and product teams to deploy models into high-availability environments.
Requirements
- Minimum of 8 years of professional experience in machine learning, specifically in deep learning and transformer architectures.
- Demonstrated history of pre-training or continually pre-training foundation models like Llama, Qwen, DeepSeek, or Mistral from scratch.
- Proficiency in multi-node, multi-GPU scaling frameworks including PyTorch FSDP, DeepSpeed, Megatron-LM, and Ray.
- Experience in LLM post-training, specifically SFT, LoRA, QLoRA, DPO, RLHF, and RLAIF.
- Strong software engineering skills in Python and C++, including experience with LLMOps and MLOps pipelines.
- Experience using AI assistants and developer agents to improve engineering workflows.
Nice to have
- Background in generative retrieval, semantic tokenization, sequence modeling of user behavior, or unifying retrieval and ranking.
Skills & tools
- Python, C++
- PyTorch FSDP, DeepSpeed, Megatron-LM, Ray
- Llama, Qwen, DeepSeek, Mistral
- SFT, LoRA, QLoRA, DPO, RLHF, RLAIF
- Mixture-of-Experts (MoE)
- MLOps, LLMOps
Practical notes
- Benefits include Term Life Insurance, comprehensive Medical Insurance, and the GrabFlex customizable benefits package.
- Leave policies include Parental leave, Birthday leave, and Love-all-Serve-all (LASA) volunteering leave.
- Access to the Grabber Assistance Programme for personal support.
- FlexWork arrangements available, including differentiated working hours.
Grab is an equal opportunity employer committed to an inclusive workplace.