3DP Agent

STL upload interface — drag and drop, no account required
Real-time 3D analysis with cognitive scan overlay
4 AI agents debate the model — Consensus Score 30/100
Light mode — full theme system propagates through Three.js scene

Inspiration

Most 3D printing tools tell you what is wrong with a model. None tell you why it will fail. I've watched prints fail at hour 6 of an 8-hour job — warped base, collapsed overhang, walls too thin to survive. The cost isn't just material. It's time, trust, and iteration cycles that shouldn't exist.

What it does

3DP Agent analyzes STL files using four specialized AI agents that debate each other:

Geometry Analyst — mesh topology, wall thickness, overhangs
Printability Scorer — weighted risk score across all dimensions
Failure Predictor — where and why it will fail
Optimization Advisor — what to change and how

They produce a Consensus Score — not one AI's opinion, but structured disagreement resolved into a verdict. A Causality tab traces failure chains from geometry decision to print outcome.

The 3D viewport animates the analysis in real time: a scan plane sweeps the model, risk markers breathe with severity, thermal fields and failure emergence appear where problems are predicted.

No account. No cloud. Runs entirely in the browser.

How I built it

React 19 + Three.js + React Three Fiber for the 3D visual layer. Multi-agent orchestration built from scratch with a debate mechanism. Claude, OpenAI, and Gemini supported via user-supplied API keys. STL parsing moved to a Web Worker for non-blocking performance on large files. Visual language system with dark/light theme that propagates through the entire Three.js scene.

Challenges

Getting four agents to produce genuinely different verdicts — not just echo each other — required careful prompt engineering per agent role. Making the 3D visual layer feel like it's thinking, not just rendering, was the hardest design problem.