Deskripsi Pekerjaan
At Nexus Intelligence Labs, we are building the foundational models that power the next generation of autonomous agents. We are a well-funded, stealth-mode startup founded by veterans from DeepMind, OpenAI, and Google Brain, operating at the intersection of reinforcement learning and large language models.
We are looking for a Senior Machine Learning Engineer to lead the architecture and scaling of our Generative AI stack. This is a high-autonomy role for a builder who wants to own the full lifecycle of model development—from data curation and distributed training to production inference optimization. You will not just be fine-tuning APIs; you will be pushing the boundaries of context windows, reasoning capabilities, and agent reliability.
We offer a hybrid environment (3 days onsite in our SoMa office) with a culture of deep technical rigor, zero bureaucracy, and direct access to massive compute resources.
Tanggung Jawab
- Design, implement, and optimize large-scale distributed training pipelines for LLMs (7B–70B+ parameters) using PyTorch FSDP / DeepSpeed / Megatron-LM.
- Architect and maintain high-throughput, low-latency inference infrastructure (vLLM / TensorRT-LLM) for real-time agentic workflows.
- Lead the development of Retrieval-Augmented Generation (RAG) systems and tool-use frameworks to ground models in proprietary knowledge bases.
- Conduct rigorous ablation studies and hyperparameter optimization to improve model reasoning, instruction following, and hallucination rates.
- Collaborate with Research Scientists to translate novel architectures (e.g., Mixture of Experts, State Space Models) into production-grade code.
- Build and maintain robust MLOps pipelines (Kubeflow / Flyte / MLflow) for experiment tracking, model versioning, and automated evaluation.
- Mentor junior engineers and establish best practices for code quality, GPU utilization, and numerical stability.
Kualifikasi
- M.S. or Ph.D. in Computer Science, Machine Learning, or equivalent practical experience (5+ years industry ML experience).
- Expert-level proficiency in Python and PyTorch; deep understanding of CUDA kernels, memory management, and GPU architecture.
- Proven track record of training or fine-tuning Large Language Models at scale (multi-node, multi-GPU clusters > 64 GPUs).
- Hands-on experience with distributed training frameworks (FSDP, DeepSpeed, JAX/Flax) and inference optimization (Quantization, Speculative Decoding, Continuous Batching).
- Strong software engineering fundamentals: Linux internals, networking (RDMA/InfiniBand), containerization (Docker/K8s), and CI/CD.
- Publications at top-tier venues (NeurIPS, ICML, ICLR, ACL) or significant open-source contributions to the LLM ecosystem (e.g., Hugging Face, vLLM, LangChain, Axolotl).
- Experience with RLHF / RLAIF / DPO alignment pipelines and synthetic data generation strategies.