Harris Oldroyd

Understand AI. Build with it. Go deeper.

I build local AI systems, run experiments, and explain what I learn — from model internals to the hardware running them.

Start learning See my experiments
Inside my local AI workstation, with two RTX 3090 graphics cardsBuilt, not just written about

The hardware behind the experiments

My local AI workstation.

48 GB of distributed VRAM, built for running models and finding out what actually works.

See the build

What do you want to explore?

No need to be an expert.

01

Understand AI

Build a mental model of tokens, attention, and how an LLM works.

02

Run models locally

Choose your hardware, pick a runtime, and build a useful local stack.

03

Compare real performance

Explore measured inference results. Keep estimates separate from evidence.

Selected work

Less hand-waving. More finding out.

Interactive explanation

See what fewer bits actually do.

Change the precision of a photo and watch the tradeoff between size and fidelity.

Try the explainer 3 min read

Inside the model

Tracing refusal through the weights.

I weakened refusal in Qwen3.5-9B and traced most of the change to one update in layer 15.

Read the experiment 7 min read

Measured performance

Why longer conversations get slower.

Memory bandwidth, the KV cache, and measured context sweeps on two RTX 3090s.

Explore the results 7 min read

Fresh from the notebook.

All AI writing