thewhitemage. avatar

thewhitemage.

0xWhiteMage

I go by the codename thewhitemage. I’m an independent builder with a background in design, exploring how to make powerful AI more practical on hardware people can own. I believe people should have the freedom to run, shape, and create with their tools; not just use them as they come.

Design shapes how I think, not just what I make. I’m drawn to ideas that don’t usually sit together, and I like questioning why things are done a certain way before accepting that they have to be. Constraints often become the starting point for a different approach.

Most of what I build begins with something I want to use myself. I learn what the work demands, experiment across disciplines, and share what proves useful. It’s how I give back while exploring how far one person’s curiosity can go.

Featured work

All work →

Focus

Models Quantisation & mixed precision, fine-tuning, speculative decoding & drafter tuning.
Agents Context management & compaction, agentic workflows, local inference tooling.
Media ComfyUI pipelines, local video generation.
Hardware NVIDIA DGX Spark, single-node local inference.

Now

Currently quantising and serving large models on a single DGX Spark, building deterministic context compaction for agent loops, and turning the process into AI content on X.

The thesis is simple: AI should enable everyone, and that starts with being able to run it yourself.