AI Engineer • Former Particle Physicist ⚛️ • Former Risk Modeler 📈

I build tools that make neural networks less mysterious.

Because I believe interpretability shouldn't stay in research papers—it belongs in production.

My work sits at the intersection of mechanistic interpretability, model visualization, and explainable AI. I'm fascinated by what happens inside modern language and vision models—and I enjoy turning those hidden processes into interactive experiments that anyone can explore.

Most of my projects are ways to look inside AI.

I'm the creator of Brainscope, an OpenAI-compatible inference server that streams live visualizations of transformer activations as a model generates text. I've also built projects exploring hidden directions in transformer weights, attention visualization, LoRA interpretability, and self-supervised vision models like JEPA.

• 🧠 Mechanistic interpretability
• 🔍 Understanding why models behave the way they do
• 📊 Interactive visualizations of neural networks
• ⚡ Practical, local-first AI
• 🚀 Open-source tools that make AI more transparent

• Transformer circuits and representation learning
• Activation steering and hidden directions
• Production tooling for interpretable AI systems

I approach AI like a physicist: observe carefully, build experiments, question assumptions, and make complex systems understandable.

My personal site is a chat with a tiny LLM running entirely in your browser, and every answer it generates renders as a real particle collision. Click the event below to fire your own question into the chamber:

Watch language models think in real time. An OpenAI-compatible server that streams live transformer activations while generating text.

Agents steering agents. One agent's mood is read off its activations and handed to another as a raw vector, no text in between. And in "the offer" mode, the receiver consents to a vector it cannot read.

A toolkit for injecting and detecting hidden behavioral directions inside transformer models.

An interpretability study exploring how tiny transformers encode writing style with LoRA adapters and attention heads.

Interactive demos explaining Meta's I-JEPA and V-JEPA models through visualizations and experiments.

Visualize attention maps and neuron activations to understand how transformer models process prompts.