Deskripsi Pekerjaan
We offer a competitive compensation package including significant equity, comprehensive health benefits, 401k matching, and a dedicated budget for compute resources and top-tier conference attendance (NeurIPS, ICML, CVPR).
Tanggung Jawab
- Design, train, and optimize large-scale deep learning models (Transformers, CNNs, GNNs) for high-throughput production environments.
- Build and maintain robust MLOps pipelines (CI/CD/CT) using Kubernetes, Kubeflow, or Vertex AI on GCP/AWS.
- Lead the end-to-end model lifecycle: data validation, experimentation tracking (MLflow/Weights & Biases), automated testing, and canary deployments.
- Collaborate with Product and Data Engineering to define feature stores and real-time inference architectures (sub-50ms latency targets).
- Conduct rigorous offline/online evaluation (A/B testing, shadow mode, interleaving) to ensure model reliability and measurable business impact.
- Mentor junior engineers and establish best practices for code quality, reproducibility, and technical documentation.
- Stay current with SOTA research (NeurIPS, ICML, ICLR) and prototype novel architectures for domain-specific problems.
Kualifikasi
- M.S. or Ph.D. in Computer Science, Machine Learning, Statistics, Mathematics, or equivalent practical experience.
- 5+ years of industry experience shipping ML models to production at scale (millions of requests/day).
- Expert proficiency in Python and modern deep learning frameworks (PyTorch preferred; TensorFlow/JAX a plus).
- Deep understanding of distributed training (FSDP, DeepSpeed, Megatron-LM), quantization, distillation, and hardware acceleration (GPU/TPU).
- Strong software engineering fundamentals: Git, Docker, Kubernetes, CI/CD (GitHub Actions/GitLab CI), and system design patterns.
- Proven track record of solving ambiguous business problems with ML solutions (Recommendation Systems, NLP/LLMs, Computer Vision, or Time-Series Forecasting).
- Excellent communication skills; ability to translate complex technical concepts to cross-functional stakeholders and leadership.