About the Role :

I'm looking for a Principal Machine Learning Engineer (Rust) to build out the infrastructure layer for an unconventional, exascale AI pre-training stack.

What we are building :

We are completely moving away from traditional, centralized error backpropagation and its massive All-Reduce network bottlenecks. Instead, we are implementing a highly distributed, gradient-free alternative inspired by Hinton's Forward-Forward paradigm.

We vertically slice model parameters across a distributed network and stream token batches horizontally. Our stack is entirely built in Rust, leveraging Burn for deep learning math, Zenoh for low-latency asynchronous data transport, and Tokio for orchestration.

To protect our IP and our clients' data, we deploy this stack using Confidential Containers (CoCo), which wraps our runtime in a hardware-encrypted micro-VM blanket (AMD SEV-SNP, Arm CCA, and Nvidia Confidential Computing) straight to the silicon layer.

By combining this stateless hub-and-edge matrix math with CoCo's zero-copy memory pinning, we can bypass standard OS serialization taxes and keep processing units at maximum compute saturation. We can pre-train a Llama 3.1 405B scale model on a fixed 460-node cluster in weeks, not months.

Our Current Status :

- Stage: Pre-revenue, validation, and core architecture design phase.

- Hardware: Planned architecture footprint (Inspur aggregators, Supermicro hubs with H100s, and Arm/Nvidia edge configurations).

- Funding/Path: We are weighing two routes: dropping this straight into a infrastructure-seed-round pitch for VC funding (the TTM numbers look incredible) OR leveraging our speed advantages to secure a signed, paying enterprise design partner while we finish the build.

What we are looking for :

You are a Principal-level Rust engineer who eats low-level systems programming for breakfast. You aren't an "AI wrapper" developer - you understand how hardware actually works. Specifically, we need deep competence in:

- Advanced Rust: Safe concurrency, async programming (Tokio), and custom memory management (zero-allocation arenas, memory pinning).

- Linux Internals & Networking: High-throughput network primitives, zero-copy architecture, gRPC/byte-streaming, and ideally RDMA.

- Infrastructural Security: Familiarity with virtualization, container runtimes (Kata), and hardware TEEs (AMD SEV, Intel TDX, Arm CCA) is a massive plus.

Total Experience : 1 - 3 years