Want to know if this job is worth applying to?
Boston, Massachusetts, United States
Onsite
Field AI is transforming how robots interact with the real world. Our R&D team, the FieldAI Research Institute (FAIRI), is based in Cambridge, MA, where we build risk-aware, field-ready AI systems that unlock general purpose intelligence for robotics.
FAIRI is looking for a DevOps Engineer to own the infrastructure our humanoid research runs on. Today that infrastructure is borrowed from Field AI's commercial platform teams and held together by researchers doing it in the margins: the humanoid monorepo has no CI, collected robot data lands in S3 and stops there, and there is no registry telling us what any given dataset actually contains.
You will be the first dedicated infrastructure hire inside FAIRI. You will build the CI/CD, data pipelines, and IaC that let a small research team ship reliably — partnering with Field AI's platform, cloud, and data-processing teams rather than rebuilding what they already run well. This is a hands-on ownership role, not a coordination role.
Stand up CI/CD for the humanoid monorepo: containerize, push to ECR, run unit tests, build, and gate on simulation system tests before promotion.
Work with the platform team's self-hosted GitHub Actions runners (ARM, AMD, CUDA, Jetson-class targets) rather than standing up parallel infrastructure.
Cut build times through change detection and remote caching — full builds are currently ~45 minutes uncached.
Build test infrastructure that lets the same test run against simple sim, Isaac Sim, or real hardware, driven over ROS 2 messages or the robot REST API.
Establish per-automation integration tests so shared-library and output-format changes cannot silently break pipelines.
Stand up and own FAIRI's Airflow stack for humanoid data processing.
Build the ingest path from robot to usable dataset: rosbag/MCAP capture, episode segmentation, format conversion, and delivery to training.
Implement data lifecycle guardrails — filtering, review-for-deletion, and retention — so idle-robot and failed-run data does not accumulate indefinitely.
Build and operate the dataset and mission registry so every dataset is attributable to a subject, session, robot, and purpose.
Support MoCapDB in production: ECS Fargate services, AWS Batch retargeting workers, RDS Postgres, and S3, integrated with FieldAI Auth.
Own FAIRI's AWS footprint as code: ECR, S3, IAM roles and cross-account trust policies, VPC and networking, Kubernetes/EKS workloads.
Close the gaps where infrastructure is not yet in code, and bring permissions changes under review.
Own compliance posture for research tooling — SOC 2 constraints on SaaS, experiment tracking, and data-sharing controls — in partnership with IT and Security.
Eliminate person-owned infrastructure: documented owners, runbooks, and access paths for every FAIRI-owned service.
Manage secrets, VPN/Tailscale access paths, and hardware-in-the-loop connectivity to robots on the floor.
Write and maintain runbooks, onboarding guides, and architecture documentation so a new engineer can test and deploy on day one rather than learning it from a teammate.
Be the interface between FAIRI and Field AI's platform, cloud, and data-processing teams — negotiating what FAIRI reuses versus owns.
Support researchers and systems engineers directly when pipelines, builds, or environments break, including live troubleshooting during demos.
Bring reproducibility discipline to research workflows: versioned configs, pinned environments, traceable runs.
You don't need every item below, but you should bring real, hands-on depth in several of them:
4+ years in DevOps, platform, infrastructure, or SRE roles
AWS in production: ECR, S3, IAM, VPC, EKS, Batch, ECS/Fargate
Infrastructure as code (Terraform, CDK, Pulumi, or equivalent) with a review-and-version discipline
CI/CD design and operation at scale — GitHub Actions strongly preferred, including self-hosted runners
Kubernetes in production, including workload scheduling and resource governance
Container tooling and build optimization: Docker, BuildKit or daemonless alternatives (Buildah, Kaniko), multi-arch builds, remote caching
Workflow orchestration — Apache Airflow or equivalent
Python, plus comfort in Bash and reading C++
Linux systems administration and networking fundamentals
Observability: logging, metrics, tracing, and alerting you actually built
Beyond the toolkit:
Comfortable being the only infra person in the room. You can take an open-ended ask and run with it without much hand-holding.
Bias toward reuse. You would rather integrate a platform team's runners than build a parallel stack, and you can negotiate that boundary well.
Strong documentation habits — you leave runbooks and processes better than you found them.
Pragmatic about research velocity. You know when to enforce rigor and when it would just slow the team down.
Infrastructure experience in robotics, autonomous vehicles, or ML research environments
ROS 2, rosbag/MCAP, or Foxglove familiarity
Large-scale data pipeline work — TB-scale sensor or video data, lifecycle and retention policy design
ML infrastructure: experiment tracking, GPU scheduling, training pipelines, simulation infrastructure (Isaac Sim / Isaac Lab)
Hardware-in-the-loop CI — running tests against physical devices from a pipeline
Compliance and audit experience: SOC 2, access reviews, data governance
Having been the first infrastructure hire on a team before
Interest in humanoid robotics and how machines learn from human movement
This role sits inside FAIRI's Research Operations function, supporting the Humanoid Program. It is a foundational hire: you will define what FAIRI's infrastructure looks like rather than inherit it, with the leverage of Field AI's existing platform teams behind you and a research team that will feel the difference immediately.
FAIRI's humanoid work runs on real deadlines with real customers among them. The infrastructure you build ships to the floor.