Six hours. That’s it. Forty-seven files, 8,000 lines of TypeScript, 15 unit tests, five serverless proxies—and a real-time 3D tracker for Artemis II spinning at 60fps in your browser.
Look, Artemis II 3D tracker demos are a dime a dozen right now. NASA launches four astronauts on Orion, first crewed trip past low Earth orbit since Apollo 17, and suddenly everyone’s got a map. But this? This one’s different. Built solo, deployed to Vercel, pulling live NASA telemetry. And yeah, an AI chatbot thrown in for kicks.
Why Does a 2026 Mission Need Your Browser Toy?
Artemis II lifted off April 1, 2026. NASA’s data? Scattered like confetti—OEM ephemeris files every four minutes, Deep Space Network XML from antennas in Goldstone, Canberra, Madrid, JPL Horizons for Moon positions. Static pages? Yawn. Someone else’s stream? Borrowed glory. This guy’s vision: interactive 3D, Lagrange interpolation for smooth 60fps curves, ground station lights blinking live.
Here’s the gem:
This post isn’t about “look what AI can do.” It’s about what happens when you give an AI agent engineering discipline instead of just a prompt.
Spot on. Every week, Twitter erupts with “I built X in 20 minutes with Claude/GPT/whatever.” Comments? Brutal. Tests? Nah. Error handling? Fingers crossed. Production-ready? Lol, no. It’s demo-ware—plausible code that crumbles when APIs hiccup or sessions timeout.
But.
This isn’t that. It’s Wrought, a pipeline the dev cooked up for Claude Code. Forces sequence: Finding → Research → Design → Blueprint → Implementation → Code Review. No skipping. Artifacts chain like Lego. Come back tomorrow? Findings Tracker spells it out—what, why, where you stalled.
Is Wrought Just Bureaucracy in Disguise?
Call it process porn if you want. But here’s my take: it’s the software engineering equivalent of Apollo’s checklists. Remember? Moon landing wasn’t astronauts winging it; it was 400,000 nerds with fault-tree analysis and redundancy obsessions. AI coding’s in its Apollo 1 phase—firey demos everywhere. Wrought? Puts guardrails.
Take the chatbot. Research stage tables it out:
| Approach | Pros | Cons | Verdict |
|—|—|—|—|
| FAQ Bot | Zero cost, instant | Rigid | Pass |
| System Prompt + LLM | Simple | API cost | Winner |
| RAG | Scales | Overkill for 3K tokens | Nope |
3,000 tokens for mission facts—crew bios, timelines, orbits. Smaller than my grocery list. RAG? That’s like building a dam for a puddle. Gemini 2.5 Flash wins on free tier (15 RPM, no card needed). Pragmatic. Not shiny.
Design? Four stacks scored. Vite + React Three Fiber tops at 8.6/10. Next.js? Hydration headaches for client-only. Vanilla Three.js? State management hell, no hot reload. Boom—decision documented.
And the result? Orion zips along trajectory, speed ticks up, Earth recedes, Moon looms. DSN antennas glow when chatting. Chatbot nails “mission length?” offline. Free-text? Curated KB via prompt. Tests pass. Errors? Handled. Down API? Graceful.
Skeptical? Me too, at first. AI speedruns scream hype. But peek at artemis-tracker-murex.vercel.app. It’s buttery. Degree-8 Lagrange polys smooth those 4-minute vectors—no jank. That’s math, not memes.
Why Does This Matter for Your Next AI Hack?
Unique angle: Wrought isn’t Claude-specific. It’s a template for any agentic workflow. Predict this: in 12 months, tools like Cursor, Aider, or even GitHub Copilot Workspace will bake in pipelines or die. Why? Because “prompt → ship → pray” scales to toys, not teams. Enterprises won’t touch untested spaghetti. This proves solo devs can punch above—production-grade in hours.
Critique the spin? Original post downplays AI wow for process win. Smart. But let’s call it: Claude Code’s still prompt-dependent. Wrought shines by chaining prompts smartly. Not autonomous agents yet—baby steps.
Dry humor time: If every AI build had this, we’d have fewer “it works on my machine” repos clogging GitHub.
The pipeline played out clean. Findings: High-severity gap, 10-day window. Research nixes overkill. Design picks stack. Blueprint? Probably wireframes, specs (cut off in original, but implied). Impl with reviews. Deploy.
Short version: Discipline beats dopamine hits.
🧬 Related Insights
- Read more: rs-trafilatura Fixes Web Scraping’s Dirty Secret: Non-Article Pages Finally Extract Right
- Read more: Seneca Preached Mercy to Nero—Got a Dagger for Thanks. Tech’s Rich Gurus, Sound Familiar?
Frequently Asked Questions
What is Wrought engineering pipeline?
It’s a structured workflow for AI coding agents like Claude: Finding, Research, Design, Blueprint, Implementation, Code Review. Enforces docs at each step—no skipping to code.
How do you build an Artemis II 3D tracker?
Grab NASA OEM ephemeris, DSN XML, JPL Horizons. Interpolate with Lagrange polys. Stack: Vite, React, Three Fiber. Pipe through Wrought for tests and polish. Deploy Vercel.
Does AI-generated code need unit tests?
Hell yes. Demos skip ‘em and break. Wrought mandates 15 here—covers interpolation, API proxies, chatbot. Production demands it.
