Deskripsi Pekerjaan
At Nexus AI Labs, we are building the next generation of foundation models that power enterprise intelligence. We are looking for a Senior Machine Learning Engineer to join our core research team in the heart of San Francisco. You will own the full lifecycle of model development—from data curation and architecture design to distributed training and production deployment.
This is a high-impact role where your work directly shapes product capabilities used by Fortune 500 companies. We value deep technical expertise, a bias for action, and the ability to translate complex research into scalable, maintainable systems.
Tanggung Jawab
- Design, implement, and optimize large-scale deep learning models (LLMs, Diffusion, Multimodal) using PyTorch and JAX.
- Build and maintain high-throughput data pipelines and training infrastructure on Kubernetes/GCP/AWS.
- Conduct rigorous experimentation: hyperparameter tuning, scaling laws analysis, and ablation studies.
- Collaborate with Research Scientists to prototype novel architectures and take them from paper to production.
- Implement model optimization techniques (quantization, distillation, compilation via TensorRT/XLA) for low-latency inference.
- Establish best practices for MLOps: CI/CD for ML, model monitoring, automated retraining, and evaluation frameworks.
- Mentor junior engineers and contribute to technical strategy and hiring.
Kualifikasi
- M.Sc. or Ph.D. in Computer Science, Machine Learning, or equivalent practical experience (5+ years industry ML engineering).
- Expert-level proficiency in Python, PyTorch/JAX, and distributed training frameworks (FSDP, DeepSpeed, Megatron-LM).
- Proven track record of training and deploying models with 10B+ parameters on GPU clusters (100+ nodes).
- Deep understanding of Transformer architecture, attention mechanisms, and optimization landscapes.
- Strong systems engineering skills: CUDA kernels, memory management, networking (NCCL, RDMA), and Linux internals.
- Experience with MLOps tooling (MLflow, Weights & Biases, Kubeflow, Argo) and cloud platforms (GCP preferred).
- Publications at top-tier venues (NeurIPS, ICML, ICLR, CVPR) or significant open-source contributions preferred.