End-to-end generative interface for AI-guided learning that keeps curiosity open.

OVERVIEW

Curiosity Engine is an AI-powered exploration tool that turns any topic into sourced, engaging reading that pulls people deeper.

A generative learning tool that turns any topic into an open-ended path to explore. Working solo, I designed the interactions, engineered the prompts, and built and deployed it as a live web app.

ROLE

Solo designer + developer

METHODS

Competitive analysis, interaction design, prompt engineering, prototyping

YEAR

2026

CHALLENGE

How might AI outputs support ongoing, non-linear learning?

Most learning online is built around getting to an answer. I wanted to see if generative experiences could turn learning into something to keep exploring, helping people follow ideas, make connections, and see where their curiosity takes them.

RESEARCH

AI can make learning easier. It can also make exploration narrower.

A small 2026 study followed people learning about nutrition with ChatGPT or Google Search over eight days. ChatGPT users started with narrower questions, rarely checked sources, and scored lower on questions that asked them to apply and evaluate what they'd learned. The researchers point to three reasons:

Offloaded selection

ChatGPT decides what information to surface, so people do less of the choosing. That choosing is part of how learning sticks.

Solution-shaped answers

ChatGPT tends to provide polished solutions rather than material to evaluate, connect, and reason through.

Narrower exploration

Search shows a range of results to compare. A chat response gives one path, which makes it easy to follow that path and miss the wider topic.

The opportunity

AI is good at making connections and surfacing ideas you wouldn't think to search for. Used well, it could open the next question instead of closing the current one.

Offloaded selection

ChatGPT decides what information to surface, so people do less of the choosing. That choosing is part of how learning sticks.

Perplexity

ChatGPT tends to hand back a finished result, like a meal plan, instead of the principles behind it. People came away with answers more than understanding.

Perplexity

Shows sources, which makes answers verifiable. But it’s built for answering questions, not exploring.

Read the study (Mittal et al., 2026)

THE PRINCIPLE

The system provides momentum. The reader sets the direction.

01

Generate momentum without taking over.

The system should be proactive enough to keep exploration moving, but not so directive that it determines where the user should go. I designed the generated content to suggest possible directions while leaving the decision of what to follow entirely to the user.

02

Let the system show its assumptions.

AI systems often make assumptions about what a user means and move forward as if those assumptions are correct. Instead of hiding that uncertainty, I made the system surface its interpretation and ask, “Is this what you meant?” before continuing.

03

Keep curiosity alive instead of closing it.

When a response gives users a definitive answer, the interaction naturally comes to an end. I redesigned the prompts to surface related questions, tensions, and possibilities so that each response could lead somewhere new.

THE PRINCIPLE

The system provides momentum. The reader sets the direction.

01

Generate momentum without taking over.

The system should be proactive enough to keep exploration moving, but not so directive that it determines where the user should go. I designed the generated content to suggest possible directions while leaving the decision of what to follow entirely to the user.

02

Let the system show its assumptions.

AI systems often make assumptions about what a user means and move forward as if those assumptions are correct. Instead of hiding that uncertainty, I made the system surface its interpretation and ask, “Is this what you meant?” before continuing.

03

Keep curiosity alive instead of closing it.

When a response gives users a definitive answer, the interaction naturally comes to an end. I redesigned the prompts to surface related questions, tensions, and possibilities so that each response could lead somewhere new.

THE PRINCIPLE

The system provides momentum. The reader sets the direction.

01

Generate momentum without taking over.

The system should be proactive enough to keep exploration moving, but not so directive that it determines where the user should go. I designed the generated content to suggest possible directions while leaving the decision of what to follow entirely to the user.

02

Let the system show its assumptions.

AI systems often make assumptions about what a user means and move forward as if those assumptions are correct. Instead of hiding that uncertainty, I made the system surface its interpretation and ask, “Is this what you meant?” before continuing.

03

Keep curiosity alive instead of closing it.

When a response gives users a definitive answer, the interaction naturally comes to an end. I redesigned the prompts to surface related questions, tensions, and possibilities so that each response could lead somewhere new.

CONTENT

Designing the curiosity loop

Every element is designed to open something, not close it. The headline draws the reader in, the body gives a range of things to pull on, and the chips point somewhere you haven’t been yet.

1

2

3

4

5

6

1

2

3

4

5

6

INTERACTION DESIGN

Interactions for a generative interface - beyond chat

When content is generated, interaction has to dissolve into behavior people already have. The reader curates by paying attention, not by navigating.

Highlight to generate

Instead of opening a prompt or tool, reading itself becomes the interaction.

Refresh content

Instead of restarting the conversation, the system offers another angle while preserving the context.

Shuffle topics

Instead of asking the user to invent a prompt, the system gives them something to notice.

REFLECTION

What I would do next.

I challenged myself to design without an input bar so exploring would feel casual. In v2, I’d add one as an option for readers who want to steer more directly, along with a chat on each card for going deeper and a layout that lets people explore across topics or drill into one. I’d also rebuild the deployment, since slow responses work against the momentum the whole tool is built on. Along the way, I learned how much an LLM’s accuracy depends on how specific I am. As my prompts got more precise, the answers I fact-checked held up.

Back to selected work