Tooling Engineer

MATS Research; Berkeley (CA); Híbrido; USD 150.000 - 190.000; Jornada completa

Ventajas ofrecidas por este puesto de trabajo

Catered meals at office

Paid staff trips and conferences

Professional development funding

17 days annual PTO

Hybrid work option

Comprehensive health insurance

Descripción de la vacante

MATS Research seeks a Tooling Engineer to build internal tooling and infrastructure that makes compute access, provisioning, budgets, and usage easier for researchers. You will contribute to GPU clusters, cloud platforms, and research tooling while collaborating with the Compute team and security to deliver scalable, maintainable solutions.

The role emphasizes owner- ship over what is built, with a preference for extending existing tools and shipping practical, user-centered software for

Formación

• 3+ years of professional software engineering experience.
• Proficiency in Python and ability to work across a web stack.
• Experience integrating with third-party APIs and building automations or data pipelines.
• Comfort working directly with users, gathering requirements and shipping iteratively.
• Familiarity with AI coding tools (e.g., Claude Code, Cursor) as part of software workflows.
• Clear written communication and documentation of builds.
• High agency and ability to scope ambiguous problems into usable solutions.
• Alignment with mission to reduce AI risk.

Responsabilidades

• Build tools and automation around compute access, provisioning, budgets, usage, and requests.
• Work with fellows and the Compute team to understand workflows and turn recurring problems into reusable tools.
• Create unified compute tooling across GPU, API, and cloud providers; extend or adopt existing tools.
• Contribute to cluster and cloud infrastructure for cost controls and observability.
• Collaborate with security to address security and access requirements.
• Develop reliable pipelines and interfaces around compute usage and spend with quality data.

Conocimientos

Python

Web stack experience

APIs integration

User collaboration

Herramientas

Claude Code

Cursor

Descripción del empleo

About the role
MATS provides research compute to its fellows across GPU providers, API providers, cloud platforms, and our own GPU cluster. The Compute team manages the platforms, budgets, and workflows that get that compute into fellows’ hands, and builds tooling and infrastructure around how researchers access and use it.

As a Tooling Engineer, you build the tools and infrastructure that make those systems easier to use, operate, and scale. Some of this is automation and interfaces around compute access, provisioning, budgets, and usage. Platform work includes giving fellows simpler ways to reach compute across many providers and giving the team better visibility and control over cost and usage. The role also includes research tooling, adopting, extending, or building the agent, evaluation, and data infrastructure fellows use in their AI safety research.

You will work closely with the Compute team, other teams across MATS, and fellows to understand where the friction is and ship useful solutions. Your work will span internal tooling and automation as well as reusable research tooling, with the mix changing as our needs evolve. You will also work with MATS’s security function, particularly where tooling and research infrastructure intersect. You will contribute to our GPU cluster and cloud infrastructure where useful; you do not need to arrive as an HPC or cloud specialist, and we expect to help you build context in these systems. This is a small team with a large surface area. You will have substantial ownership over what gets built, with a strong preference for adopting and extending existing tools over building from scratch when they fit.

You might be:

• A full-stack or backend engineer who has built internal tools or platforms for technical users and enjoys the product side as much as the code.
• A data or platform engineer who has built pipelines and integrations against third-party APIs and wants to work more closely with research.
• A research engineer or technical generalist from an AI lab, startup, or safety organization who ships quickly and wants to build infrastructure that other researchers use.

If you do not fit perfectly into one of these profiles but believe your skills are a strong match for the responsibilities, we are still excited for you to apply.

Key responsibilities
Tooling and automation

• Build tools and automation around compute access, provisioning, budgets, usage, and requests, and automate recurring workflows for teams across MATS.
• Work directly with fellows and the Compute team to understand workflows, identify operational bottlenecks, and turn recurring problems into reusable tools.

Compute platforms and infrastructure

• Build simpler, more unified ways to access and manage compute across the GPU, API, and cloud providers we support, adopting and extending existing tools where they fit, writing a thin MATS layer where that is enough, and building from scratch only where nothing else serves.
• Contribute where useful to our cluster and cloud infrastructure, and to the controls and observability around cost, access, and usage.
• Work closely with MATS’s security function on the security and access requirements of our research infrastructure and tooling.

Data and observability

• Build reliable pipelines and interfaces around compute usage, spending, capacity, and operational state across the platforms we support.
• Make that information useful to fellows, the Compute team, and organizational planning and decision-making, with clear provenance and strong standards for data quality.

Research tooling

• Adopt, extend, or build tools for AI safety research, including agent and evaluation infrastructure (e.g. Inspect, PETRI) and data infrastructure that helps researchers capture, organize, analyze, and learn from their work.
• Work with fellows and residents to understand their research workflows and build reusable tooling that can accelerate research across projects.

Across all of this, you keep what you build maintainable, documented, and deployable by others on the team, and you iterate with real users rather than in isolation.

Examples of projects you might work on

• Compute access and automation. Automate common provisioning, top-up, and request workflows across supported platforms, with human approval where spend or policy requires it.
• Unified compute tooling. Improve how fellows access compute across providers and how the team monitors usage, spend, and capacity.
• Research tooling. Adopt or build tooling that helps fellows run and evaluate AI safety research, such as deploying an agent-evaluation framework for use on our platforms.
• Research data infrastructure. Build systems that make research outputs and data easier to capture, organize, query, and reuse.

Essential qualifications and skills

• 3+ years of professional software engineering experience, with production code you have shipped and maintained.
• Strong software engineering skills, including proficiency in Python, and the ability to work across a web stack or willingness to become effective across it.
• Experience integrating with third-party APIs and building automations or data pipelines against them.
• Comfort working directly with users, gathering requirements iteratively, shipping small, and refining from real usage.
• Effective use of modern AI coding tools and agents (e.g. Claude Code, Cursor) as part of your software-development workflow.
• Clear written communication; you document what you build.
• High agency; you can take an ambiguous problem, scope it, and drive it to something people use.
• Alignment with our mission of reducing risk from the development of advanced AI.

Desirable qualifications and skills

We expect especially strong applicants to have experience in at least one of the following:

• Building internal tools, dashboards, or platforms for researchers, engineers, or other technical users.
• Research or ML tooling, such as agent infrastructure, evaluation systems, experiment tooling, research-data pipelines, or other researcher-facing engineering.
• React and TypeScript.
• Cloud administration and cost control on GCP or AWS; familiarity with GPU providers, inference routers, and how researchers use them.
• Data engineering, including ETL, analytics storage, and reconciliation or metrics pipelines.
• Linux, HPC, or ML infrastructure (e.g. Slurm clusters). Useful, not required; we expect to help you build context here.
• Prior work in an AI safety organization, fellowship, or research lab.

About the Compensation

• The essential skills listed above contain the basic requirements, and align with the lower end of this range.
• MATS provides in-office, catered lunches and dinners to employees on workdays.
• Paid work trips, including staff retreats, business trips, and relevant conferences.
• Funding and support for professional development.
• 17 days of annual paid time off for Berkeley-based employees.
• Flexibility for hybrid work (but not for fully remote work).
• Collaborative and intellectually stimulating work environment.
• Medical, dental, vision, and life insurance.
• Both Roth 401(k) and Traditional 401(k) for US-based employees.

Working hours and location

40 hours per week. Successful candidates can expect to spend most of their time working in person from our main office in Berkeley, California. We are open to hybrid working arrangements for exceptional candidates.

MATS is committed to fostering a diverse and inclusive work environment at the forefront of our field. We encourage applications from individuals of all backgrounds and experiences.