Deskripsi Pekerjaan
Nexus AI Labs is building the foundational infrastructure for the next generation of generative AI applications. We are a well-funded, early-stage startup founded by ex-researchers from top-tier labs (Google Brain, FAIR, OpenAI) seeking a Senior Machine Learning Engineer to own the lifecycle of our core model training and deployment pipelines.
You will operate at the intersection of cutting-edge research and high-scale engineering, designing systems that train massive models efficiently and serve them with low latency to millions of users. This is a high-autonomy, high-impact role for an engineer who wants to define the technical trajectory of a category-defining company.
Why Join Us?
- Technical Depth: Work on novel architecture optimization, distributed training at scale, and advanced quantization/compilation techniques.
- Ownership: End-to-end responsibility for the modeling stack—from data curation to production inference.
- Compensation: Top-of-market salary, generous equity (ISO options), 401k matching, and comprehensive health/wellness stipends.
- Culture: Low bureaucracy, high trust. We optimize for speed of iteration and technical excellence.
Tanggung Jawab
- Design, implement, and optimize distributed training pipelines for Large Language Models (LLMs) and multimodal models across GPU clusters (100s–1000s of GPUs).
- Architect and maintain our MLOps platform: experiment tracking, model registry, automated evaluation pipelines, and CI/CD for ML artifacts.
- Lead model optimization efforts including quantization (GPTQ, AWQ), distillation, compilation (TensorRT, Triton, XLA), and speculative decoding for low-latency serving.
- Collaborate with Research Scientists to translate novel modeling ideas (e.g., MoE, RWKV, State Space Models) into production-grade, scalable code.
- Build robust data processing pipelines (Petabyte-scale) for pre-training, instruction tuning, and RLHF/RLAIF alignment using Spark/Ray/DataProc.
- Establish rigorous evaluation frameworks (automated benchmarks, human preference testing, red-teaming) to guarantee model quality and safety.
- Mentor junior engineers and define best practices for code quality, reproducibility, and GPU utilization efficiency.
- Contribute to open-source tooling and publish findings at top-tier conferences (NeurIPS, ICML, ICLR, MLSys).
Kualifikasi
- Education: MS/PhD in Computer Science, Machine Learning, or equivalent practical experience (5+ years industry ML engineering).
- Frameworks: Expert-level proficiency in PyTorch (internals, autograd, custom kernels) and/or JAX; strong familiarity with Hugging Face ecosystem (Transformers, Accelerate, PEFT, TRL).
- Distributed Systems: Deep hands-on experience with distributed training paradigms (FSDP, DeepSpeed, Megatron-LM, tensor/pipeline/data parallelism) and cluster orchestration (Slurm, Kubernetes, Ray).
- Serving & Optimization: Proven track record optimizing inference latency/throughput using TensorRT-LLM, vLLM, Triton Inference Server, or custom CUDA kernels.
- Data Engineering: Fluency in large-scale data processing (Spark, Polars, Ray Data) and data quality/cleaning for LLM pre-training corpora.
- Software Craftsmanship: Production-grade Python (type hints, testing, profiling), Rust/C++ for performance-critical paths, and Git/GitLab/GitHub CI workflows.
- Cloud/Native: Experience deploying ML workloads on AWS (p4d/p5 instances, EFA, FSx for Lustre) or GCP (TPU pods, A3 VMs).
- Communication: Ability to articulate complex technical trade-offs to both technical and non-technical stakeholders.