LoadLab

Inspiration

I wanted to build something that did more than tell students if a bridge passed or failed. LoadLab helps students understand why it failed and use that evidence to improve their design.

My goal was to use AI as a coach without letting it take over the thinking or building.

What it does

LoadLab is an AI-guided bridge lab for students in grades 6–8.

Students predict what will happen, test a bridge with moving vehicles, inspect stress and movement, make a change, and test again. The AI coach gives clues based on their actual results, but the student still decides what to build.

How I built it

I built LoadLab with Next.js, React, TypeScript, Three.js, and React Three Fiber.

I created a deterministic truss solver that calculates forces, movement, and failure. GPT-5.6 looks at selected solver evidence and the student’s actions, then gives a short coaching response.

The solver is always the source of truth. The AI cannot change the bridge, invent engineering results, or give the student a finished solution.

I also used Vitest and Playwright to test the solver, student flow, AI rules, accessibility, and fallback experiences.

How I used Codex

I used Codex throughout the entire project, from choosing the idea to building and testing the final experience.

Codex helped me move much faster by turning my ideas into product specifications, helping write the solver and interface, connecting GPT-5.6, running tests, and checking the experience in the browser. It also helped me find smaller issues that were easy to miss, like clues that technically worked but did not give students an obvious next step.

I still made the main product, engineering, and design decisions. I decided that students should propose and build their own solutions, the solver should be trusted over the AI, and the project should focus on one strong 10–15 minute learning experience.

GPT-5.6 contributed in two ways. Through Codex, it helped me work through product, engineering, and design problems. Inside LoadLab, it powers the evidence-based coach. The code decides what happened to the bridge, while GPT-5.6 helps the student think about why.

Built With

• codex
• gpt5.6
• three.js