What's It Made Of?

Inspiration

We interact with hundreds of objects every day without knowing what they are made of. Chemistry often feels disconnected from real life with symbols and formulas. I wanted to make that invisible world easy to grasp. ChatGPT can explain what something contains, but seeing and exploring the molecular structures inside it creates a much stronger understanding.

What it does

What’s It Made Of? turns a photo of an everyday object into interactive 3D molecules. The Matter Scanner 3000 identifies the object, estimates a few representative substances inside it, and reveals the molecules as neon 3D structures. Users can rotate the structures, open individual atoms, and build custom molecules from the periodic table in the Atom Lab. Signed-in users can also save previous scans and collect newly discovered molecules in their Discovery Vault. Results are clearly presented as educational estimates (a photograph cannot measure an object’s exact chemical composition).

How I built it

I built the project through an iterative collaboration with Codex in the ChatGPT desktop app, using GPT-5.6 Sol at medium reasoning with full access to the project. I supplied the product vision, scientific and UX requirements, real-device testing, and final decisions. Codex proposed and implemented the architecture, features, fixes, regression tests, and documentation. I continuously steered the work with observed failures and new requirements, using /goal for longer implementation tasks. All development took place during the build week.

Challenges I ran into

The hardest problem was turning AI output into structures that could be rendered reliably. A valid-looking JSON response could still contain disconnected atoms, impossible bonds, misleading crystal structures, or a formula that did not match the model. I built several validation layers, a focused repair attempt, and a fail-closed renderer so bad structures never silently reach the user. Mobile browsers introduced another set of challenges: camera inputs, audio restrictions, touch timing, safe areas and more. Repeated iPhone testing was essential.

Accomplishments that I'm proud of

I am most proud that the app already has active users: my family and nephews.

I am happy that this became a complete product rather than only an AI demo. Users can photograph almost any recognizable object, watch it transform into molecular structures, manipulate those structures directly, inspect individual atoms, build molecules, and save discoveries across sessions.

Most importantly, I actually enjoy using the app, and it continues to surprise me what things actually contain!

What I learned

GPT-5.6 is excellent at recognizing objects and proposing meaningful explanations. Traditional code is better at enforcing formulas, graph consistency, supported bonding rules, and renderer safety. Combining the two produced a result that was both flexible and reliable.

I started using /goal with Codex, this is a real game changer.

What's next for What's It Made Of?

I will expand molecule building to support more of the periodic table in the Atom Lab. I will improve the atom visualization. I will open source it, since I think it is especially useful for classrooms and young learners.

My biggest hope is that the project reaches teachers and eventually gets used in real classrooms.

Built With

• codex
• openai
• openai-api
• postgresql
• railway
• react
• three.js
• typescript