Deskripsi Pekerjaan
Shape the Future of Generative AI
At Nexus Intelligence, we are building the foundational models that power the next generation of enterprise automation. We are looking for a Senior Machine Learning Engineer to join our core research team in the heart of San Francisco. This is a high-impact role where you will own the full lifecycle of large-scale model training, from data curation and architecture design to distributed training optimization and production deployment.
Why Nexus?
- ✦Compute Freedom: Unlimited access to H100/A100 clusters via strategic cloud partnerships.
- ✦Open Culture: Publish at top venues (NeurIPS, ICML, ICLR); strong open-source contribution encouraged.
- ✦Ownership: End-to-end responsibility for model capabilities, not just feature engineering.
- ✦Compensation: Top-of-market salary, significant equity (ISO), 401k match, and premium healthcare.
If you thrive in ambiguity, obsess over training stability, and want to define how LLMs reason, we want to talk to you.
Tanggung Jawab
- Design, implement, and scale novel model architectures (Transformers, Mixture of Experts, State Space Models) for pre-training and fine-tuning on massive datasets.
- Optimize distributed training workflows using PyTorch FSDP, DeepSpeed, or Megatron-LM across thousands of GPUs, targeting maximum MFU (Model FLOPs Utilization).
- Build and maintain high-performance data pipelines for multi-modal datasets (text, code, vision) using Spark, Ray, or custom Rust/Go tooling.
- Conduct rigorous ablation studies and hyperparameter scaling law experiments to guide architectural decisions.
- Collaborate with Research Scientists to translate theoretical breakthroughs into production-grade training recipes.
- Develop custom kernels (CUDA/Triton) for attention mechanisms, normalization layers, and quantization-aware training.
- Establish best practices for model evaluation, red-teaming, and alignment (RLHF/RLAIF/DPO) pipelines.
- Mentor junior engineers and contribute to the technical direction of the ML platform team.
Kualifikasi
- 5+ years of industry experience in Deep Learning, with at least 2+ years specifically focused on Large Language Model (LLM) training or large-scale distributed systems.
- Expert proficiency in Python and PyTorch (or JAX); strong systems programming skills in C++, Rust, or Go for performance-critical paths.
- Deep understanding of distributed training paradigms: Data Parallelism, Tensor Parallelism, Pipeline Parallelism, Sequence Parallelism, and ZeRO optimization.
- Proven track record of debugging training instability (loss spikes, gradient explosion, NaNs) and optimizing GPU kernel utilization (NCU/NSight profiling).
- Experience with MLOps/LLMOps tooling: MLflow, Weights & Biases, Kubeflow, or internal platform development for experiment tracking and model registry.
- Strong grasp of Transformer architecture internals (RoPE, FlashAttention, Grouped-Query Attention, Rotary Embeddings).
- Advanced degree (MS/PhD) in Computer Science, Machine Learning, Mathematics, or equivalent practical experience with publications in top-tier conferences.
- Excellent communication skills; ability to articulate complex technical trade-offs to both technical and non-technical stakeholders.