Software Engineer, Compute Infrastructure (US)
Dedalus Labs • San Francisco, CA
Company
Dedalus Labs
Annual salary
$275k – $350k/yr
Location
San Francisco, CA
Listed
27d ago
- Experience:
- 5+ years
- Workplace:
- On-site in San Francisco. Relocation support is available. Travel to datacenters is required when deployment and operations call for it.
- Equity:
- 1.0% - 1.5%
Open to people who live in the US, Canada or the UK
We can only put candidates forward for this role from those three countries for now. See the contract AI training jobs open in more countries.
On-site in San Francisco, CA.
Send us your LinkedIn and CV. If your experience fits, we'll introduce you to the recruiter filling this role.
Apply for a referral →The recruiter emails you before anything happens and may suggest other jobs that suit you better.
What you'll do
- Bring GPU capacity online by working with datacenter operators, hardware vendors and onsite technicians to turn delivered servers into usable capacity. This covers racking, power, cooling, cabling, networking and remote management. The goal is to have GPUs online by the end of November 2026.
- Take ownership of the first bare-metal GPU service, which customers access over SSH. Make practical, incremental progress with a simple service before building a larger software platform.
- Provision, automate, debug and operate production Kubernetes and server fleets, including cluster and node lifecycle, networking, storage, capacity, upgrades, observability and incident response.
- Run Linux systems administration and SRE work: user and SSH access, system services, package and configuration management, patching, diagnostics, reliability and recovery. Automate provisioning, OS images, firmware, drivers and configuration step by step.
- Set up acceptance tests that check GPUs, memory, storage and network performance before machines join the fleet. Build monitoring and diagnostics that detect, isolate and recover unhealthy nodes.
- Coordinate repairs and replacements with vendors, lead incident investigations, and turn recurring failures into software fixes. Make upgrades, maintenance and recovery safe and easy to understand.
- As needs emerge, grow the compute control plane: APIs, controllers, scheduling, and integration with Dedalus Machines and our Rust virtual machine monitor. This includes how GPUs are assigned, isolated, exposed to guests and safely reclaimed.
- Work closely with the CTO and founders on architecture decisions. Help hire and mentor engineers as the team grows.
- Choose existing tools where they fit and build only what the product needs. Write tests, deployment workflows, documentation and operating procedures that other engineers can maintain.
- Use AI coding tools to speed up your work. You must be able to explain the design, verify its behavior and diagnose failures yourself. Logistics: Full-time and on-site in San Francisco. Travel to datacenters is required when deployment and operations call for it. Relocation support is available. Salary is $275,000, $350,000 plus 1.0%, 1.5% equity. Meals and office benefits are included.
What you need
- We need a senior or staff-level systems engineer who has personally brought bare-metal GPU capacity, or similar compute, online and kept it running reliably for real users. That includes working with datacenters and vendors. There's no fixed years-of-experience requirement. Exceptional new graduates will be considered only if they show substantial, relevant hands-on work. You must be skilled at Kubernetes in practice: provisioning, automating, debugging and operating a production cluster, and writing and understanding its code and configuration. You should be comfortable owning hands-on deployment and production operations as the first engineer and technical lead of a new compute infrastructure team. You should also have a track record of measurable reliability and customer results. Bonus points for Slurm scheduling and accounting, Environment Modules (module load), shared Python/CUDA environments, configuration management, infrastructure as code, modern toolchains such as uv, NVIDIA GPU systems, InfiniBand/RoCE, or virtualization (KVM, QEMU, device passthrough).
Common questions
How do I apply for the Software Engineer, Compute Infrastructure (US) role at Dedalus Labs? +
Use the Apply for a referral button on this page to send us your LinkedIn and CV. If your experience fits, we'll introduce you to the recruiter filling this role. They'll email you to check you're interested, then put you forward for this job or others that suit you better. It's free.
What does this Dedalus Labs role pay? +
The listing gives $275k – $350k/yr.
Is this role remote? +
The listing gives the location as San Francisco, CA. On-site in San Francisco. Relocation support is available. Travel to datacenters is required when deployment and operations call for it.
Software Engineer interview prep
Questions Software Engineer interviewers often ask. They aren't taken from this listing.
How do you approach designing a system when the requirements are still likely to change?
I design around the parts of the problem I'm confident are stable, like the core data model, and keep the parts most likely to change, like specific business rules, isolated behind clear interfaces. Over-engineering flexibility everywhere just adds complexity to parts of the system that were never going to change.
What's your process for debugging an issue that only reproduces in production, not locally?
I add targeted logging around the suspected area and compare production data or configuration against my local setup, since an issue that's environment-specific usually traces back to a real difference between the two rather than a fundamentally different bug. I avoid making speculative changes without first confirming where the divergence is.
How do you decide when a piece of code needs a test versus when it's low-risk enough to skip?
I prioritize tests for logic with real business consequences if it breaks, or code that's likely to be touched again later by someone unfamiliar with it. Simple, rarely-changed glue code gets less coverage, since the cost of a thorough test there often exceeds the risk it's protecting against.
How do you evaluate a proposed technical solution's trade-offs, like performance versus maintainability?
I start from the actual constraint that matters for this specific system, whether that's genuinely performance-critical or whether maintainability is the bigger long-term cost, rather than defaulting to whichever trade-off I personally prefer. Optimizing for performance the system doesn't need just adds complexity nobody benefits from.
Related roles
Browse all referral roles
Salaried roles at startups, AI companies and established businesses, all filled by referral. One application covers every role.
See all referral roles →