190,000 Star Systems and a Million Planets: How I'm Building GalaxyAtlas with AI Agents

Description and use

GalaxyAtlas is a browser‑based 3D galaxy with 190,000 star systems and 1,000,000 procedurally generated planets, built using AI models and optimized for ordinary computers.

Problem

Developing a large‑scale, browser‑based 3D galaxy with realistic procedural planets while keeping performance acceptable on ordinary computers.

Outcome

GalaxyAtlas demonstrates that a complex, procedurally generated galaxy can be built and maintained using AI‑driven engineering, achieving a playable experience entirely within a web browser.

Workflow

• 1. Validate concept with GPT Sites to create a quick prototype using HYG catalog data.
• 2. Expand prototype to include procedural galaxy generation (disk radius 15.3 kpc, pitch angle 13°, bar orientation 27°).
• 3. Implement lazy generation, LOD, resource reuse, and scene‑based resource release to keep GPU memory and draw calls low.
• 4. Use deterministic seed‑based procedural generation for planets and environments to allow repeatable exploration.
• 5. Integrate Three.js for rendering, applying atmospheric effects, fog, particles, and lighting while monitoring performance.
• 6. Iterate with AI models (GPT‑6 Astra) to generate code, fix bugs, and refine architecture.
• 7. Publish the final product as a browser link without a separate client.

Evidence

I managed to build a browser‑based atlas using data on more than 109,000 real stars from the HYG catalog and a procedural model of the galaxy.

Source

GalaxyAtlas runs entirely in the browser, displays ~190,000 star systems, allows navigation, landing, and exploration of procedurally generated planets, and uses AI‑driven development to manage complexity.

Source

Three.js became the foundation of GalaxyAtlas's 3D rendering.

Source