Senior Machine Learning Engineer

Role Overview

Join a core model engineering team as a Senior Machine Learning Engineer focused on building and scaling large language models (10B–100B+ parameters). This role bridges applied research and production-grade engineering, with ownership across the training lifecycle and system performance.

What You’ll Own

• Foundation Model Engineering (End-to-End)
  • Implement large-scale pre-training, SFT, and alignment pipelines.
  • Optimize architectures and training strategies using scaling laws and product goals.
  • Improve performance, reasoning capability, and training efficiency.
• Distributed Training & Performance Optimization
  • Design and optimize multi-node GPU distributed training (A100/H100/B200).
  • Use Data / Tensor / Pipeline / Sequence Parallelism.
  • Maximize MFU and cluster efficiency.
  • Improve stability, fault tolerance, and monitoring.
• High-Throughput Data Systems
  • Build TB–PB scale data pipelines.
  • Implement ingestion, cleaning, deduplication (MinHash/LSH), safety filtering, and PII removal.
  • Support multimodal strategies, synthetic data generation, and curriculum learning.
• Applied LLM Research Implementation
  • Productionize alignment methods: RLHF, DPO, KTO.
  • Work with Mixture-of-Experts (MoE) and routing optimization.
  • Improve reasoning, math, and coding performance.
  • Build/extend agent and tool-calling systems.
• Engineering Excellence
  • Maintain strong code/system design standards.
  • Identify and eliminate performance bottlenecks.
  • Own major components end-to-end.

Requirements

• Education/Depth: MS/PhD in CS/AI/Math or equivalent practical experience.
• Hands-on engineering optimizing large-scale deep learning systems.
• Strong Transformer knowledge (e.g., RoPE, FlashAttention, SwiGLU).
• Experience with modern LLMs (open-source or proprietary).
• Proficiency in PyTorch or JAX.
• Distributed training frameworks such as Megatron-LM, DeepSpeed, or FSDP.
• Knowledge of 3D parallelism and ZeRO optimization.
• Training on large GPU clusters (100+ GPUs preferred).
• Familiarity with InfiniBand, RDMA, and storage I/O optimization.
• Ability to debug large distributed training runs.

Nice to Have

• Open-source contributions in LLM ecosystem.
• Experience with agentic systems or multi-step reasoning frameworks.
• CUDA/Triton kernel optimization.
• Published research or major production LLM deployments.