Careers
Build the layer between the GPU and the invoice.
We are a small founding team with an AI first repository, hardware from NVIDIA and AMD on the bench, and buyers who want receipts. The first hires shape the company.
How we work
Ship the number, then talk about it.AI writes the first draft. A person owns the result.If it is not measured, it is an opinion.Small team. Real hardware. No theatre.
We started because a working change was turned away for how it was written, not for what it did. So we judge work by what it does. Bring a result and the measurement behind it and you will be heard, whoever you are and however you made it.
Agents draft, you decide
The repository is AI first. Agents write the first pass, certified benchmarks gate the merge, and a person decides what ships.
Hardware you can touch
DGX Spark and Strix Halo are on the bench today, from NVIDIA and AMD. Hopper and Blackwell are the next campaigns.
The number is the argument
Every claim ships with the measurement behind it, including the rungs we lost on the way. Nobody wins a review by seniority.
Early enough to matter
The team is small enough that the first hires decide how the company works, not only what it builds.
The fast track
Open a pull request.
The repository is public. Pick an issue, land a change, and mention it in your note. It is the introduction we read first, and it tells us more than an interview does.
The first hires
Four roles we are building toward
Send a note and a link to something you built to careers@metrale.com. Code beats resumes. A pull request against the engine beats both.
Kernel Engineer Location by agreement · Engine
CUDA and Rust. Attention, MoE, GDN and quantized GEMM kernels per hardware target, with a certified benchmark on every merge. Hopper and Blackwell are the next campaigns.
The work
- Write and tune kernels for one hardware target at a time
- Land every change behind a certified benchmark
- Open the Hopper and Blackwell campaigns
What gets our attention
A kernel you wrote, and the number it moved.
Founding Platform and Distributed Systems Engineer Location by agreement · Control
The control plane. Routing, rollout, autoscaling, repair and policy across mixed NVIDIA and AMD fleets, in Rust, on Kubernetes, in the customer’s account and ours.
The work
- Build routing, rollout, autoscaling, repair and policy in Rust
- Run it on Kubernetes, in the customer’s account and in ours
- Keep the control plane off the inference path
What gets our attention
A scheduler or control plane you ran in production.
Founding Infrastructure and Forward Deployed Engineer Location by agreement, travel · Customers
You stand up the pilot, run the ladder on the customer’s hardware, and hand them the receipt. The person the customer calls, and the person who tells engineering what broke.
The work
- Stand up the pilot on the customer’s hardware
- Run the ladder and hand over the receipt
- Bring what broke back to engineering
What gets our attention
A deployment you carried from the first call to production.
Strategic Product and Marketing Advisor Part time · Go to market
Former product, sales or GTM leadership at an inference or GPU infrastructure company. Monthly working sessions, pricing and positioning review, select enterprise calls.
The work
- Monthly working sessions with the founders
- Review pricing and positioning
- Join select enterprise calls
What gets our attention
Product, sales or go to market leadership at an inference or GPU infrastructure company.
Next step
Do not see your role?
Send the note anyway. A pull request against the engine is the best introduction there is.