An AI reading tool designed to feel like a rabbit hole, not a search result.

Context
2026 • Personal concept
Role
Solo designer + developer
Methods
Competitive analysis, prompt engineering, prototyping
Focus
Interaction design, UX for generative interfaces
Single Work Image

overview

Inspired by the feeling of a well-edited magazine, following curiosity without knowing where it leads. Built solo to learn what it means to design for generative output.

Competitive Landscape

AI interfaces I learned from

Few frameworks exist yet for this kind of interface, but a few products solved pieces of it.

Process Icon
Perplexity

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

Process Icon
NotebookLM

Showed that generative AI doesn't have to be chat. But it's limited to documents you upload.

Process Icon
Claude / ChatGPT

Works well when you know what to ask. Less useful when you're just curious.

Process Icon
Curiosity Engine

Borrows from all three: visible sources, a format other than chat, and suggestions for where to go next.

Prompt Engineering

Problems from early sessions, and how I fixed them.

The early copy was safe, so it said nothing

Headlines were vague. Body copy qualified everything before getting to the point. I fixed this with one prompt test: does this sentence sound like something a person would actually say out loud? Every card now leads with the punchline.

Wrong

"Sleep is essential for memory consolidation, and researchers continue to explore the complex mechanisms involved in how the brain processes information during rest cycles."

Right

"Your brain chemically paralyzes your whole body every night so you don't act out your dreams."

The engine would confidently pick the wrong meaning of an ambiguous topic.

Type "Mercury" and you might get the planet, the god, or Freddie, and whichever it picked, it would commit. I fixed this by adding a pre-flight pass that catches ambiguous prompts and asks one short follow-up before generating. Now the app asks what you meant instead of guessing.

Highlighting a concept often led somewhere adjacent

Highlight "developmental trauma" and the next card might pivot to a person or event instead of defining the concept. I rewrote the branch prompt to treat the highlight as the subject itself, so the next card explains the thing you tapped. Reading now follows the user's curiosity instead of redirecting it.

Prompt Architecture

Creating a prompt framework that hooks

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

1
Domain tag
Shows where you are in the topic, like a room label in a museum.
4
Body
The specifics are the interaction surface. Readers branch by highlighting, so every card needs at least three things worth grabbing.
2
Header
Opens with the wildest true version of the fact. The first reaction should be "wait, what?"
3
Subheader
Earns the trust the headline spent. Plainly explains what the headline means.
5
Sources
Real references behind every claim, so the rabbit hole stays trustworthy.
6
"Explore further" suggestions
Three directions that imply genuinely different sessions, not three zoom-ins. A five-lens framework forces the spread.

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

Highlighting is already a reading behavior. The interaction lives inside the content.

Refresh button

A different angle on the same topic. What you've built survives.

Shuffle topics

For users with no topic in mind. Shuffled chips remove the blank-page problem.

Interface Design

Two ways to navigate your curiosity

The list is memory. The tree is possibility.

List view

Shows every card you've read, in order. A quick answer to "what did I just cover?"

Tree view

Shows every fork, including the paths you didn't take. Jump back to any fork and branch differently. The session never ends, it just pauses.

Reflection

Through this process, I learned that designing UX for AI means designing the output just as much as the interface. I found that interactions need to feel like things people already do, or the content can't do its job. At its best, learning starts with a question. Building this was an exploration of how AI might generate more questions, not just more answers.