Member of Technical Staff
SF Tensor
Member of Technical Staff
SF Tensor
Hardware-agnostic AI and HPC software stack
We firmly believe that the future of AI depends on the unglamorous: rethinking and rebuilding the stack, all the way down. From the hardware underneath it to the compiler targeting it and the cloud running it. Right now those three things fight each other and that friction shows up as a tax on every researcher trying to build something ambitious. We're here to axe that tax and make compute faster, cheaper and more available. When we succeed, compute will be portable enough that "which cloud, which chip" stop being something you worry about.
To achieve this, we are building our Kernel Optimizer, which takes code and finds its fastest possible form for whatever vendor and cluster topology you point it at, automatically, as well as the Model Foundry which manages the runs, makes research easier and moves workloads across clouds and chips as prices and availability ship, instead of leaving you locked into whatever vendor you signed with first.
We're backed by Susa Ventures, Y Combinator, along with some great funds and angels including Max Mullen and Paul Graham, as well as founders and executives at Neuralink, Notion and AMD. We're looking for researchers, engineers and organizations who agree with the basic premise: you don't get the next leap in AI without a leap in compute first.
We build the fastest GPU compiler in the world. Most compilers have to preserve correctness at every transform, constraining how far they can search, while we prove correctness at the end instead, allowing us to search a far wider space, with agents, with RL, with anything that works and still guarantee the result. It's why we hold #1 on NVIDIA's own kernel benchmark across hundreds of production kernels.
Speed at the kernel layer is only worth what we do with it though, so we're hiring a Member of Technical Staff for Product Engineering to own the surface that the engine reaches the world through.
That surface is Model Foundry and it takes a researcher from idea to a running experiment in one commit or one message. When a researcher needs to see the perplexity curve from last night's run or to try an experiment with a new annealing rate, Foundry writes the config, queues the job on whatever silicon makes sense that week, then stream the logs back and versions the whole thing so it can be reproduced six months later. You'll own it end to end from the interfaces researchers live in all day to the services behind them, making sure the software doesn't just work, but the experience feels right and is enjoyable even when using it 16 hours a day for months on end.
The users are close and the loop is short. Foundry is used by both our customers, our internal research team as well as our forward-deployed researchers to train models every hour of the day, so your users either sit next to you or a Slack message away. The data you're processing is not gentle: you'll be processing telemetry from thousands of GPUs across NVIDIA, AMD, TPU and Trainium from runs that last for weeks to months parsing logs and charts that don't stop and making sure all of the data is accessible to user as well as their agents.
You'll take full ownership of product areas in Model Foundry: scoping, designing, building and iterating on them
You'll build clean, fast and polished web interfaces with NextJS that feel almost native
You'll design the surfaces where researchers live: run dashboards, live log streaming, experiment comparison, eval results, cluster and storage views
You'll build the conversational and commit-driven entry points into Foundry, where a Slack message or commit become a training run
You'll obsess over the small things, from loading states to smooth animations and the edge cases that turn a good product into an exceptional one
You'll work directly with our forward-deployed researcher and our own training team. They're your users, they sit next to you and their feedback loop is measured in minutes
You'll make smart product tradeoffs with little hand-holding, which means knowing when to ship fast and when to spend extra time polishing
Someone with proven experience shipping complete products, ideally something you can show us
Someone with real taste and a sharp eye for design and motion, which means you notice when an animation eases awkwardly, when padding feels inconsistent or something just feels cheap
Someone with solid backend chips, which means you can build APIs and troubleshoot production issues
Someone with enough infrastructure comfort to handle AWS, containers and Kubernetes (you don't need to be a DevOps wizard but shouldn't be scared of a Dockerfile either)
Someone with experience working with Bun or an eagerness to dive in
Someone with a background in developer tools, infrastructure products or building for technical users
Someone with any exposure to ML training workflows or tools like Weights & Biases, Ray or Slurm
Someone comfortable using Figma to mock up things yourself when needed
Most infrastructure companies treat their interface as an afterthought and their users can tell. The engineering and engine behind our product is as good as it gets and the product on top deserves to be just as good. You'll be defining what researchers at frontier labs feel when they train a model and are staring at logs and curves for hours on end with the goal of making the research experience as enjoyable as possible and let the infrastructure get out of the way and fade into the background.
We're a small team operating at frontier scale. We pre-trained foundation models on 4,000 AMD GPUs as a team of three, designed and brought up GB300 NVL72 clusters and designed a TOP500 supercomputer.
We believe that hard problems get solved in person and most of our work happens at our office in San Francisco. We offer relocation assistance and, where possible, we'd like you here as often as possible.
The base salary range for this full-time position is $225,000-$275,000, plus meaningful equity and benefits.
SF Tensor builds software and infrastructure for AI and HPC to reduce the infrastructure burden. It offers Emma Lang, a hardware-agnostic language that runs efficiently across GPUs and TPUs without rewriting code, plus the SF Tensor Stack with Kernel Optimizer and Elastic Cloud. Kernel Optimizer transforms models by simulating hardware topology to reach efficient forms, often outperforming hand-tuned code; Elastic Cloud finds the best-priced hardware across clouds and coordinates training from one to thousands of GPUs. Its goal is to make high-performance computing fast, affordable, and portable so AI teams can focus on research instead of infrastructure.
Explore more opportunities at SF Tensor