Make the first step obvious
Students should type, speak, or upload a problem without stopping to learn the interface.
Design proposal for MathGPT
A math tutor that helps you learn, not just answer.
MathGPT asked us to reimagine their AI math tutor. We designed a full concept, end to end, and proposed it to their team. It never shipped, but it's a clear look at how we approach a hard, feature-dense product problem.
- Client
- MathGPT
- Engagement
- Design proposal
- Scope
- Strategy, UX/UI, mobile
Answering a question is easy. Supporting learning is harder.
An AI math tool can produce an answer in seconds. The harder part is helping a student understand what happened, practice the idea, and come back to it later.
MathGPT had a lot to fit into one product: chat, video explanations, study guides, graphs, flashcards, practice tests, classes. All without turning into a crowded dashboard.
Connect the learning tools
Every explanation should lead naturally into practice, revision, and a saved class context.
Work beyond desktop
The core experience had to stay focused and useful on a much smaller mobile screen.
The existing MathGPT experience.
One workspace we designed, from question to understanding.
The prompt sits at the center. Every other tool is arranged around the work a student is already doing.
We kept the current question in focus while classes, recent activity, and learning formats stayed one click away. A dark, low-distraction canvas gave the content room to breathe.
Each format, video, practice, study, graphs, flashcards, got its own color. That was the wayfinding, so we didn't need more menus.
- Product strategy
- UX architecture
- Interface design
- AI interaction design
- Responsive web
- Mobile experience
Keep the conversation and the learning context together.
Uploaded material, the student’s request, the generated explanation, and the next prompt all stay visible in one continuous workspace.
Three rounds toward the final concept.
Every version tightened the layout and sharpened the visual system, until the interface matched how a student actually studies.
Every learning format, one visual system.
Video explanations, class organization, and the surrounding UI shell all share the same patterns, so nothing feels like a bolted-on feature.
A finished concept, not yet a shipped product.
We delivered a complete design concept, from the core workspace to mobile, and proposed it to the MathGPT team as a direction for the product’s next stage.
MathGPT hasn’t moved this design into production. We’re sharing it here because it’s a clear example of how we approach a hard, feature-dense product problem, not a claim about what MathGPT users see today.
One place to learn
Questions, explanations, practice, study material, and classes were all designed as part of the same screen.
AI you don't have to learn first
Clear prompts, familiar controls, and an obvious next step, so a student could just start.
Same model on every screen
The desktop workspace and the mobile app were designed on the same patterns, so nothing would feel like a different product.
The same workspace, built for a much smaller screen.
More work
Explore other stories.
Building an AI product people have to trust?
Let's make it feel useful, not just clever.
Bring the model and the messy first build. We'll help you find the right place to start.