__tl;dr: low-level/kernel engineer, $150k-300k / year, part-time possible, remote __ Many AI companies are now focusing on low-level development — because for the first time in history, models can do something there, which means specialists are needed to verify and refine that something ⚙️

Last week, Anthropic posted a take-home challenge for kernel developers: optimizing tree traversal on a custom processor.

The baseline result is 147k cycles, Claude Opus 4.5 scores a hundred times less. If you can do better, you get a fast track. Did you see it? Maybe you even participated? 👀

If you're not ready to relocate even for Anthropic — you can work remotely with them and other top labs at Preference Model. The team is building RL environments that help models perform less reward-hacking and handle complex tasks better. The company's founders are from Anthropic and DatologyAI, and they've secured $16M from Andreessen Horowitz + SignalFire. They are especially looking for people now who can create environments for training LLMs on fast, low-level code 🖥

This role is for you if you understand: ⚡️ low-level optimization — CUDA/HIP, memory and caches, assembler, C++, OR ⚡️ compilers and JIT — Triton, JAX/XLA, LLVM, MLIR, OR ⚡️ ML inference — vLLM, PyTorch internals, NCCL.

🔘You also know Python and English at C1+ level.

In the previous post, you guys hated the contractor format — we were surprised!

We genuinely think it's awesome for many cases. If, for example, you're currently doing a PhD — you can take on 20 hours a week. Want to try without a long commitment — no problem.

And if you want full-time — it will come out to $150k-300k per year remotely, with the possibility to transition to a full-time employee if you both like each other and you want to move to SF.

Your take-home from Lunapark — write to Fedor @owlkov and tell him about your low-level development experience 🌟