Machine Learning Systems & Infrastructure Engineer

About the role

SpAItial is building the next generation of world models that understand physics and geometry, enabling generative 3D AI for robotics, AR/VR, gaming and cinema. As a Machine Learning Systems & Infrastructure Engineer you will own the end‑to‑end pipelines that turn raw real‑world data into trained models and production endpoints, working closely with a small research‑focused team.

Key responsibilities

• Design, implement and operate scalable training stacks, dataset loaders, checkpointing and experiment orchestration for large diffusion models.
• Improve distributed training performance and stability using PyTorch DDP/FSDP, NCCL and pre‑emption‑safe sharded checkpoints.
• Build end‑to‑end Python pipelines that ingest, clean and version petabyte‑scale training data from third‑party sources.
• Operate ML workflow orchestration tools (Kubeflow Pipelines, Airflow), GPU schedulers (Volcano, Slurm) and experiment trackers (MLflow, Weights & Biases).
• Containerize workloads with Docker and Kubernetes, maintain IaC via Terraform, and set up CI/CD pipelines with self‑hosted GPU runners.

Required profile

• 3+ years of production‑quality Python development in a large, multi‑author codebase.
• Hands‑on experience with modern ML training stacks (PyTorch, DDP/FSDP) and debugging distributed GPU jobs.
• Proven ability to ship complex data pipelines at scale, including handling rate‑limited or undocumented APIs.
• Practical knowledge of cloud platforms (AWS, GCP or Azure), object storage, IAM and cost‑aware engineering.
• Experience with containers (Docker, Kubernetes) and infrastructure‑as‑code tools such as Terraform.

Required skills

• Python
• PyTorch (DDP/FSDP)
• Docker & Kubernetes
• Terraform
• SQL (Postgres, BigQuery, Snowflake, SQLite)
• Cloud platforms (AWS, GCP, Azure)
• Kubeflow Pipelines / MLflow
• Prometheus & Grafana