Deskripsi Pekerjaan
At Nexus Intelligence Labs, we are architecting the next generation of foundation models for enterprise automation. We are seeking a Senior Machine Learning Engineer to lead the lifecycle of large-scale distributed training, optimization, and deployment of LLMs and multimodal systems.
You will operate at the intersection of research and production, collaborating with Research Scientists to translate novel architectures into robust, scalable services running on massive GPU clusters. This is a high-autonomy role for an engineer who thrives on solving complex systems challenges and pushing the boundaries of model performance and efficiency.
Why Join Us?
- Compute Access: Dedicated access to 1000+ H100/A100 clusters.
- Open Culture: Strong commitment to open-source contributions and publishing at top-tier venues (NeurIPS, ICML, ICLR).
- Impact: Your code directly powers AI agents used by Fortune 500 companies.
Tanggung Jawab
- Design, implement, and optimize distributed training pipelines (FSDP, DeepSpeed, Megatron-LM) for models 7B–70B+ parameters.
- Architect high-throughput inference serving stacks (vLLM, TensorRT-LLM, Triton) optimizing for latency, throughput, and cost.
- Lead model quantization (GPTQ, AWQ, FP8) and compilation (XLA, TVM, TensorRT) strategies for edge and cloud deployment.
- Build and maintain MLOps infrastructure: experiment tracking, model registry, CI/CD for ML, and automated eval pipelines.
- Collaborate with Research to productionize novel architectures (MoE, RWKV, Mamba, Retrieval-Augmented models).
- Perform rigorous GPU kernel profiling (NCU, Nsight) and write custom CUDA/Triton kernels for bottlenecks.
- Establish best practices for data curation, synthetic data generation, and curriculum learning at scale.
- Mentor junior engineers and drive technical roadmap for the ML Platform team.
Kualifikasi
- M.S. or Ph.D. in Computer Science, Machine Learning, or equivalent practical experience (5+ years industry).
- Expert-level Python and PyTorch/JAX; strong proficiency in C++ and CUDA programming.
- Proven track record of training and deploying LLMs or large-scale vision models on multi-node GPU clusters.
- Deep understanding of Transformer architecture internals, attention optimization (FlashAttention, PagedAttention), and memory management.
- Extensive experience with distributed systems concepts: MPI/NCCL, gradient accumulation, pipeline/tensor parallelism.
- Hands-on expertise with Kubernetes (Kubeflow, KServe), Slurm, or Ray for orchestration.
- Strong publication record or significant open-source contributions to ML frameworks (PyTorch, HuggingFace, vLLM, JAX).
- Excellent communication skills; ability to distill complex technical trade-offs to cross-functional stakeholders.