Escalation engineer at Microsoft working on AI and IoT. Spending my nights and weekends going deeper into systems programming because I want to understand how things actually work — not just use them.
Over the last few months I've built a CPU matmul kernel from scratch with SIMD and cache blocking, a tensor engine, and a forward pass neural network, all in pure Rust, zero dependencies. Recently went down the async Rust rabbit hole and built a mini runtime from scratch — raw vtables, custom wakers, no Tokio. Dense and rewarding.
Now moving into GPU programming. Same approach, understand it from the ground up before using abstractions.
Long term I'm moving toward ML inference engineering, the intersection of low-level systems, GPU programming, and the infrastructure that makes LLMs actually run fast.
- Rust — SIMD, async internals, systems programming
- CUDA — learning GPU programming through PMPP
- Azure AI / IoT — day job, escalation engineering
- CUDA matmul kernel
- Inference scheduler in Rust