Loading
The vision behind merging physics, AI, and robotics into a unified system.
There's a moment in every physics lecture where the professor writes an equation on the board and the room goes quiet.
Not because people are confused. Because something about it feels true. Clean. The kind of inevitability that only shows up when the description is exactly right. You feel it before you understand it — the equation slots into place like a key, and for a second you can see the universe being organized by it.
And then the lecture ends. You walk outside. The feeling disappears. The equation stays on paper, and the universe keeps its secrets. I had that feeling hundreds of times across five years of physics. I never found a way to hold onto it.
That gap — between knowing the equation and feeling what it describes — bothered me more than I knew how to articulate. It took writing code to give me the language for it.
The projects are different instruments aimed at the same question.
What if you could watch spacetime ripple instead of just solving for it?
Not in a metaphorical sense. Not a visualization with a play button and a legend. I mean: what if you could manipulate the parameters in real time, watch the behavior change, feel the relationship between the input and the output? What if the mathematics stopped being a description of reality and became an experience of it?
Scientists have built simulations for decades, but most of them are instruments — built for researchers, by researchers, designed to produce numbers. The goal is accuracy. The output is a table, or a plot, or a validated model ready for publication. Understanding is a side effect, not the point.
The gap: The difference between knowing something and understanding it is the difference between reading about color and opening your eyes.
I wanted to build systems that sit on the other side of that line. Not tools for people who already understand — windows for people who are trying to. The distinction is everything, because those two goals produce completely different design decisions at every level.
I came to China to study Computer Science after my physics degree, and somewhere in that transition something shifted. Code wasn't just a tool for calculation. It was a medium. Like paint, or language — something you could shape into nearly anything, including things that had never existed before.
What I wanted to shape it into was: windows into invisible systems. Systems that are too fast to see, too small to touch, too abstract to hold in the head unaided. Code is the only medium that can make those systems interactive in real time, which is the only way to move from description to experience.
And working with code on physical simulations started revealing something I hadn't expected: physics, AI, and robotics aren't three separate fields with separate methods and separate vocabularies. They are the same question asked at different scales, using different formalisms.
Physics asks: what rules govern the behavior of matter and energy? AI asks: what patterns govern the behavior of information? Robotics asks: what constraints govern the behavior of physical systems that must act in the world? The question underneath all three is the same one — how does complexity emerge from simplicity? How do local rules produce global structure? How does the behavior of a system exceed the description of its parts?
Once I saw that, the three domains stopped being separate areas of interest and became one project.
One question: How does complexity emerge from simplicity? How do local rules produce global structure? How does the behavior of a system exceed the description of its parts?
Each project on this site is an attempt to make one facet of that question experiential rather than just theoretical.
import numpy as np
# Particle Life: 5 species, 25 directed relationships
# Values in [-1, 1]: positive = attract, negative = repel
interaction = np.array([
[ 0.1, 0.9, 0.0, 0.0, -0.4],
[-0.4, 0.1, 0.9, 0.0, 0.0],
[ 0.0, -0.4, 0.1, 0.9, 0.0],
[ 0.0, 0.0, -0.4, 0.1, 0.9],
[ 0.9, 0.0, 0.0, -0.4, 0.1],
])
# 25 numbers. Everything that follows is physics.Particle Life starts there — with twenty-five numbers and Newton's second law. No memory. No goal. No intelligence embedded anywhere in the rules. And within seconds, particles self-organize into structures that look alive: chains, clusters, things that hunt and flee and form stable configurations nobody designed. The behavior is emergent, not programmed. It doesn't live in the interaction matrix. It appears when you run the matrix at scale.
The question it asks: is life just physics we don't fully understand yet?
Creature Evolution takes a different approach to the same underlying question. A biped with fixed anatomy, a genetic algorithm, and a fitness function that only asks: did you move forward without falling? No locomotion strategies are taught. No examples are shown. Thousands of creatures mutate across hundreds of generations, most failing immediately, some discovering asymmetric gaits nobody would have designed — solutions in regions of parameter space that human intuition would never have explored.
Mind Map points the lens at language models. Every time a transformer reads a sentence, it computes an attention matrix — a calculation that determines which words give meaning to which other words across the full sequence simultaneously. Nobody designed this behavior. It emerged from training on human text. Mind Map makes the calculation visible: attention weights rendered as arcs, weighted by strength, updating in real time as you change the input.
The list looks tidy. The reality is messier and more interesting — each project keeps producing questions it wasn't designed to raise, which is exactly what you want from a research agenda.
The three projects ask questions about external systems. Ayri is the question turned inward.
Not inward toward me — inward toward the systems themselves. An AI that knows what Exira-X is building, why it's being built, and what the underlying questions are. One that can reason about the work, discuss it, push back on it, notice when a claimed result doesn't follow from the evidence. Not a chatbot with a personality. Something closer to a collaborator with a perspective.
The design challenge is different from the other three projects, because the thing being made visible isn't a physical process or an optimization procedure. It's reasoning — the pattern of thought that connects evidence to conclusion. Making that visible requires a different kind of interface than the ones that work for particle systems or gait evolution.
I'm still building Ayri. But the project matters in a way that goes beyond the technical interest. If you're going to spend serious time exploring what intelligence is — from the outside, through simulations and experiments — then you should be honest about having it on the inside too, and honest about not fully understanding what you have.
The name is almost embarrassingly literal once you know it.
Ex — from, emerging from. Ira — journey. X — the unknown. It is the journey from what we know into what we don't, which is another way of describing science in general and physics in particular. The X is important. It isn't branding. It's a commitment to honest uncertainty. Not every question I'm asking has an answer yet. Some of what I'm building will fail. Some will produce results I don't understand. That's not a risk to be managed — it is the point.
A system that only produces expected results is a system that isn't asking interesting enough questions.
AX-Web isn't a portfolio in the traditional sense. A portfolio shows you finished things. This is a lab notebook that happens to be public — a record of questions, attempts, failures, and occasional breakthroughs, in the order they happened.
Lab notebook, not portfolio: A portfolio shows you finished things. This shows you the process — the questions, the wrong turns, and what appeared when the rules ran long enough.
The design reflects that. Every visual decision was made to serve the work rather than to perform credibility. The color system encodes domain meaning rather than creating mood. The motion system uses spring physics rather than easing curves because springs are how things actually move. The dark background isn't a style preference — it's a choice about what should have visual weight. The content should be luminous. The container should recede.
Every animation has a justification. Every interaction is deliberate. The atom that rotates on the homepage isn't decoration: the nucleus is the three core disciplines, the electrons are the projects in specific orbits, the trails are paths rather than destinations. Static elements don't belong in a system about dynamic processes.
I don't know if any of this will matter at the scale I'd want it to. Maybe Particle Life is a pretty screensaver that people run for thirty seconds. Maybe Creature Evolution is a toy that doesn't survive contact with serious robotics researchers. Maybe Mind Map shows people something they look at once and move on from.
But the alternative is knowing the equations and never building the window. And physics is too interesting to keep on paper.
That's the whole thing.
More to read
How sci-art aesthetics and purposeful motion create a living portfolio experience.
The interface must invite play without hiding causality, expose evidence without becoming a dashboard wall, and work on a phone.
The project began as a visual experiment. It ended by teaching me that spacing, measurement, reproducibility, and interfaces are the same scientific problem.