// Software Developer & Systems Engineer — Oregon, USA

KERF BIT

RODNEY VANMARTER

I fix what others gave up on.
In metal. In code. In anything.

Scroll to explore
35+ Years Problem-Solving
32K Lines of C++ Written
567+ Tests Passing

THE FORGE
DOESN'T CARE
WHAT YOU
THOUGHT WAS
IMPOSSIBLE.

A forge takes raw, unworkable material — something heavy, stubborn, resistant — and through heat, pressure, and skill, makes it into something precise and strong. That's not a metaphor I chose. It's how I've always worked.

Before I wrote neural networks, I machined steel. Before I built REST APIs, I rebuilt industrial furnaces and designed the preventative maintenance schedules to keep them running. I've spent decades operating at the intersection of hardware and software, metal and code — and I've learned that the hardest problems aren't unsolvable. They just haven't met the right person yet.

I'm that person.

⚙️

Systems Thinker

I don't fix symptoms. I trace failures to their root, whether that's a worn bearing, a memory leak, or a flawed architecture. Everything is a system. Every system has a solution.

⌨️

C++ Native

I build close to the metal. C++17, manual memory management, transformer architectures without frameworks. I know what's happening underneath because I wrote it.

🔥

Built to Last

567+ tests. Full CI/CD. Docker deployment. Documented architecture. I don't ship prototypes and call them products. I build things meant to run in production.

Featured Work

// Flagship Project — C++ · AI · Systems

ADAI

Advanced Deep Learning AI — Built from scratch. No frameworks. No shortcuts.

A production-grade transformer neural network engine, written entirely in C++17. Encoder-decoder architecture. Multi-head attention. BPE tokenization. Adam/AdamW optimization. KV cache delivering 2–3x inference speedup. A full REST API server. Docker deployment. Cross-compiled Windows executables from Linux.

Most people learn AI by calling a Python library. I learned it by implementing scaled dot-product attention by hand and making it fast.

C++17 CMake Transformers REST API Docker Google Test CI/CD Python Shell
32,500 Lines of Code
567+ Test Suites
100% Pass Rate
79 Source Files
2–3× Inference Speedup

I have rebuilt machines declared dead. Replicated parts no longer manufactured. Written AI from first principles. The problem doesn't matter. What matters is that I don't stop.

GOT A PROBLEM NO ONE'S SOLVED?