Founding Product Engineer

Project Crete, Inc.; San Francisco (CA); On-site; USD 180,000 - 320,000; Full time

Job summary

Project Crete is building personal AI hardware in San Francisco. We are hiring a Founding AI/ML Engineer to make frontier-quality intelligence run beautifully on consumer hardware and to own local inference end to end.

You will ship ML systems, optimize across CPU/GPU/NPU targets, and build the edge runtime with CUDA, Metal, Vulkan, and NPU SDKs. Strong knowledge of PyTorch and low-level performance work is essential.

Qualifications

• Shipped ML systems people actually used.
• Proficient with PyTorch and low-level performance work.
• Portfolio over pedigree; code-level design skills.

Responsibilities

• Own local inference end to end — quantization, distillation, speculative decoding, memory-efficient serving.
• Build the agent runtime: tool use, memory, on-device orchestration.
• Profile and optimize models across CPU, GPU and NPU targets.
• Work in PyTorch and low-level performance areas to push hardware efficiency.
• Develop the edge runtime — schedulers, drivers, acceleration paths across CUDA, Metal, Vulkan and NPU SDKs.
• Drive end-to-end performance: latency, memory, thermals.
• Design and implement the app surfaces — desktop client, web, onboarding and developer experience.
• Collaborate with product/design to shape local AI on consumer hardware.

Skills

C/C++/Rust

TypeScript

Embedded systems

Performance optimization

Tools

PyTorch

CUDA

Metal

Vulkan

NPU SDKs

Job description

We're a small team in San Francisco, building personal AIhardware — it runs your agents locally, on hardware you own.Private by physics, not by promise. We hire rarely and deliberately.

How we work

• Humans stay in the loop.AI accelerates the work; architecture, judgment and taste stay human.We don't ship prompt-generated foundations.
• Hardware and software, one stack.Software alone is no moat. Lasting products are atoms and bitsdesigned together — device, install, service, ecosystem.
• Three generations deep.Gen one works, gen two refines, gen three scales. We accept earlyniches and early losses; stopping is the only failure.
• Candor over comfort.Hard truths early, surfaced kindly. Polite lies rot a system.Key calls are made decisively; everything else is yours to own.
• Story before specs.We start from why it matters to a real person — press releasefirst, features second. Technology serves the narrative, not the reverse.
• Ethics as a feature.No addiction loops, no dark patterns, no burnout culture.We build for people's lives — including our own.

Founding AI/ML Engineer

San Francisco · Full-time

Make frontier-quality intelligence run beautifully on consumer hardware.

• Own local inference end to end — quantization, distillation, speculative decoding, memory-efficient serving.
• Build the agent runtime: tool use, memory, on-device orchestration.
• Profile and squeeze models across CPU, GPU and NPU targets.
• You've shipped ML systems people actually used, and you're at home in PyTorch and in low-level performance work alike.
• Build the edge runtime — schedulers, drivers, acceleration paths across CUDA, Metal, Vulkan and NPU SDKs.
• Performance engineering end to end: latency, memory, thermals.
• Fluent in C, C++ or Rust; kernel- or embedded-adjacent backgrounds very welcome.
• Build the app surfaces — desktop client, web, onboarding and the developer experience.
• Full-stack TypeScript, with taste for fast, quiet, native-feeling software.
• You sweat the details users feel, and skip the ones they don't.

Define how private, local AI looks and feels.

• Own product design end to end — flows, UI systems, brand expression.
• Portfolio over pedigree; motion and visual craft are a plus.
• You can design in code, or right beside it.