Deskripsi Pekerjaan
Nexus Intelligence is at the forefront of applied AI, building enterprise-grade generative platforms that redefine how Fortune 500 companies interact with unstructured data. We are a well-funded, Series B startup backed by top-tier VCs (Sequoia, a16z), operating with a flat hierarchy and a high-trust, asynchronous culture.
We are looking for a Senior Machine Learning Engineer to own the full lifecycle of our core foundation models—from data curation and distributed training to low-latency serving and rigorous evaluation. You will not just implement papers; you will architect the infrastructure that allows us to iterate faster than the competition.
If you thrive in ambiguity, obsess over GPU utilization metrics, and want your code to directly power the next generation of AI agents, this is your runway.
Tanggung Jawab
- Design, train, and optimize large-scale transformer models (LLMs/ Diffusion) on multi-node GPU clusters (100+ H100s/A100s).
- Build and maintain a robust MLOps platform (Kubeflow, MLflow, custom tooling) for experiment tracking, feature stores, and automated CI/CD/CT pipelines.
- Implement advanced training techniques: FSDP/DeepSpeed, mixed precision, activation checkpointing, and model parallelism.
- Develop high-throughput, low-latency inference engines using TensorRT-LLM, vLLM, or custom CUDA kernels.
- Curate and version massive datasets (PB-scale) using Spark/Ray; design synthetic data generation pipelines for alignment.
- Collaborate with Product and Research to define evaluation benchmarks (automated & human) that correlate with real-world user value.
- Mentor junior engineers; establish coding standards, testing rigor, and documentation practices for the ML team.
- Stay current with ArXiv publications; rapidly prototype and productionize promising architectures (e.g., Mixture of Experts, State Space Models).
Kualifikasi
- 5+ years professional experience in ML Engineering; 2+ years specifically in Large Language Model training/serving.
- Expert-level Python; strong systems programming skills in C++ or Rust for performance-critical paths.
- Deep hands-on expertise with PyTorch (internals a plus), JAX, or TensorFlow; proficiency with Hugging Face ecosystem (Transformers, Accelerate, PEFT).
- Proven track record optimizing distributed training (Megatron-LM, DeepSpeed, FSDP) and inference (vLLM, TGI, TensorRT-LLM).
- Strong grasp of GPU architecture (NVIDIA Hopper/Ampere), CUDA programming, and kernel optimization.
- Experience building data-intensive applications with Ray, Spark, or Flink; familiarity with Kubernetes operators for ML workloads.
- BS/MS/PhD in Computer Science, Mathematics, or equivalent practical experience; publications at top-tier venues (NeurIPS, ICML, ICLR) are a strong signal.
- Excellent communication skills; ability to distill complex technical trade-offs for cross-functional stakeholders.