$ cat ~/about.txt
I'm PhilSupertramp, a software engineer with a mathematics background. I work at the intersection of machine learning, numerical computing, and backend infrastructure — usually by building things to understand them.
$ ls -la ~/projects/featured
A curated set of research projects and developer tools. They illustrate the problems I enjoy working on — not a claim that every project is production-ready.
The question: How far can we train neural networks with integer arithmetic, rather than only quantizing them after training?
A Rust experiment with fixed-point representations, explicit scaling, stochastic rounding, and numerical correctness work.
Experimental; actively evolving
Explore repository →The question: Can distributed learning experiments be simple to run, inspect, and extend?
Python infrastructure for federated training with client/server execution, dataset sharding, persistent work queues, and FedAvg aggregation.
Explore repository →The problem: API behavior regresses, while the scenarios that originally mattered disappear from memory.
A smoke-testing toolkit with declarative cases and a proxy-assisted workflow for recording request/response examples.
Explore repository →The problem: Exposing different generative model tasks through one consistent API is harder than it sounds.
Experiments with a unified serving interface across text, speech, and image model pipelines, including PyTorch and ONNX workflows.
Explore repository →$ cat ~/background.md
My background is in mathematics and software engineering. For my B.Sc. thesis, I investigated machine-learning methods for localizing and classifying insects in images. In 2023 I taught an introductory machine-learning class at Berliner Hochschule für Technik.
I'm interested in the seam between theory and implementation: why an algorithm works, how it behaves numerically, and what it takes to run reliably in a real system.
$ ls ~/archive
More experiments live on GitHub.
Interested in machine learning systems, Rust, infrastructure, or open-source development? I'm happy to exchange ideas about the work.