The Full Build Story

How I Built a Solar System in a Weekend

From Jupiter to Pluto in 48 hours with AI — the complete story of Solar System Explorer, how it was built, and what it means.

By Kyle Ewing Build time: 1 Weekend Versions: v1 – v10 Systems: 10

The Spark

It started with a memory.

Sometime in the mid-1980s, I played a game that let me pilot a spacecraft through the solar system in first person. The physics weren't perfect. The graphics were primitive. But something about the experience stuck — the sense of scale, the silence, the feeling of actually being out there.

Decades later, nothing on the market had captured that feeling the way I remembered it. Flight simulators, space games, NASA visualizations — none of them were quite right. So one Saturday morning, instead of looking for something that didn't exist, I decided to build it.

That decision took about 30 seconds. What followed was one of the most absorbing weekends I can remember — and the project you're reading about now.

The Approach

No WebGL experience. No development team.
Just conversation and iteration.

I used Claude and Fable — Anthropic's agentic coding tool during its introductory period — as my primary development partner. Three.js handled the 3D rendering. Cloudflare Workers provided zero-cost, unlimited-bandwidth hosting.

I'm not a developer. I lead major technology programs for a living — 8-figure enterprise implementations, infrastructure transformations, organizational change. I know what it takes to build software at scale. This was nothing like that. And that's the point.

The process that emerged was closer to engineering than vibe coding. Before fixing anything, I measured it first. Wrong diagnoses cost more time than wrong code. When something broke, I described what I observed — not what I thought the cause was — and let Claude reason from there.

One pattern that dramatically accelerated the build: parallel AI workers. Rather than one linear conversation, I ran multiple Claude sessions on separate subsystems simultaneously — orbital mechanics in one, shader development in another, UI polish in a third. The outputs got reviewed and merged. Build time per version dropped significantly.

The tradeoff: parallel workers require careful review. AI can confidently produce code that conflicts with decisions made in another session. The human has to hold the architecture in their head and catch the contradictions. AI is a force multiplier — but judgment still lives with you.

Every prompt was an architectural specification, not just an instruction. The difference matters enormously at scale.

The Build

Version by version.

Ten versions. One weekend. Here's what got built and in what order.

v1

Jupiter — The First Working System

Core Jupiter system with 6 camera modes, real orbital mechanics for the Galilean moons, and spatial audio. The moment Jupiter appeared in the browser for the first time — textured, lit, rotating — was the moment I knew this was going to work. Everything after was refinement.

v2–v4

Polish, GeoSync, UI Redesign

Orbit insertion mechanics, altitude control, procedural detail shaders that add infinite-resolution zoom, and GeoSync orbit — where the camera locks to the planet's rotation and the surface sweeps beneath you. UI was rebuilt from scratch. The oval Jupiter bug was diagnosed and fixed. Source code was lost before first commit and recovered. The 21-module Vite build mystery was solved.

v5

Earth + Moon — City Lights, Auroras, Apollo Sites

Earth required the most texture layers of any body: cloud layer, city lights on the night side, auroral glow at the poles, surface detail. Six Apollo landing sites are marked with surface coordinates. The Moon includes highlands, mare, and crater detail from NASA's Lunar Reconnaissance Orbiter data.

v6

Mars — Olympus Mons, Valles Marineris, Dust Storms

Mars elevation data from NASA's MOLA mission drives the procedural terrain. Olympus Mons — the tallest volcano in the solar system — is visible from orbit. Valles Marineris stretches across the equator. Dust storm events occur procedurally. Phobos and Deimos orbit accurately.

v7

Saturn — Ring System, Ring Shadows, 9 Moons

The hardest build of the project. Saturn's ring system required accurate geometry, particle density variation, and self-shadowing — rings casting shadows onto the planet surface and vice versa. Titan's thick atmosphere glows orange. Enceladus erupts geysers from its south pole visible from 200 km. Nine moons total, all on accurate orbits.

v8

Mercury, Venus, Uranus, Neptune — All 8 Planets

The final four inner/outer planets completed the set. Mercury's heavily cratered surface from MESSENGER imagery. Venus rendered through its thick cloud layers. Uranus and Neptune with their distinctive color signatures and ring systems. All 8 planets now in a single simulator.

v9

The Sun — Photosphere, Corona, Solar Flares

The Sun was the most technically complex single body — a dynamic object rather than a static texture. Photosphere granulation, coronal streamers, active sunspot regions, solar flares, and prominences all rendered procedurally. The scale correction required to make the Sun feel appropriately vast while keeping the planets navigable was its own engineering problem.

v10

Pluto + Charon — New Horizons Imagery, Mordor Macula

The final system. Pluto's surface textures derived from New Horizons mission imagery — Tombaugh Regio (the heart-shaped nitrogen plains), the dark equatorial band known as Mordor Macula, blue nitrogen haze in the thin atmosphere. Charon orbits as a true binary system — both bodies orbiting their shared center of mass. The solar system was complete.

12h
Keyboard Time
1 Weekend
Calendar Time
10
Systems
40+
Moons
$0
Hosting Cost
1
AI Partner

What I Learned

Four things that actually matter.

Lesson 01

Measure before fixing.

Wrong diagnoses cost more time than wrong code. Before asking Claude to fix anything, I described what I observed — not what I assumed the cause was. This single habit eliminated most dead ends.

Lesson 02

Parallel workers cut time — but require review.

Running multiple AI sessions simultaneously accelerated the build dramatically. But AI workers don't know about each other. The human holds the architecture and catches the contradictions. That role doesn't go away.

Lesson 03

v1 decisions echo through v10.

The architectural choices made in the first version — coordinate systems, shader approach, camera model — shaped every version that followed. Time spent on v1 foundations pays compounding returns. Shortcuts at the start are expensive later.

Lesson 04

AI is a force multiplier, not a replacement.

AI can lead you into problems faster than you'd find them alone, if you're not careful. It's not a substitute for judgment, domain knowledge, or clear thinking. With those things, it is genuinely transformative.

The Bigger Point

This is what The Doodle Principle looks like in practice.

I wrote The Doodle Principle about what becomes possible when curiosity, creativity, and AI intersect. This project is the live demonstration. A curiosity became a memory. A memory became a question. A question became a weekend. A weekend became a solar system.

Three years ago, building something like this required a team, a budget, years of specialized experience, and probably a game engine license. Today it requires curiosity, passion, and a willingness to try. The tools have finally caught up to the imagination.

That's not just true for me. It's true for anyone with an idea they didn't know how to build. The barrier isn't technical skill anymore. It's the decision to start.

Read The Doodle Principle

Ready to explore?

The solar system is waiting. No download, no install.