General-purpose LLM
- Generates language
Can also classify intent, call tools, and return structured decisions.
“Check the ride service’s pet policy.”
Or a structured choice: Pet travelThe app reads the answer or uses the choice.
An interface that starts
with what you need.
Describe your situation. JEV makes decisions within the product’s options. The app uses them to show a relevant screen or ask a question.
Same situation, different next screen
“I’m bringing my dog and a suitcase.”
Fixed flow
Where to?
PickupDowntown
DestinationAirport
Set your pickup and destination.
Continue
Start with the trip.
Intent-driven flow
Where to?
PickupDowntown
DestinationAirport
Set your pickup and destination.
Continue
Start with the same trip.
Trip entry is illustrated. JEV’s decisions are live.
Try your own situationWhere JEV fits
“I’m bringing my dog.”
JEV
- Selects from defined options
Screens, actions, or states the product allows.
- Choice: Pet travelConfidence + choice probabilitiesThe app uses the decision to show pet guidance.
you define the choices
Both can make choices. General-purpose LLMs support open-ended generation and reasoning. JEV specializes in fast, bounded decisions with typed outputs and confidence signals.
A valid choice can still be wrong, especially with ambiguous, incomplete, or unfamiliar input. The app can use confidence to proceed, ask a question, or refer to a person.
Illustrated outputs. This demo can ask for clarification; it has no human handoff.Use cases
Each product defines its screens and choices. JEV’s decisions and confidence help the app choose a next step. Try rideshare; the other examples are illustrations.
Rideshare
Live demo“I’m bringing my dog and a suitcase.”
Pet-travel guidancePolicy · Carrier requirements
Luggage summary1 suitcase · Already supplied
Unclear request? The demo asks you to choose a supported need.
Try the rideshare demoE-commerce
Illustration“I need a gift by Friday, under $50.”
- Budget + delivery date
- Choose a defined screen
Delivery filter
Gift shortlist
If more context is needed: “Which Friday, and where should it arrive?”
Fixed: Browse → Filter → Product detailsIllustration only. No live integration.
Onboarding
Illustration“I’m moving my team from another tool.”
- Migration + team setup
- Choose a defined screen
Import checklist
Invite teammates
If more context is needed: “Which tool are you moving from?”
Fixed: Welcome → Profile → Generic tourIllustration only. No live integration.
Insights
Illustration“Why did repeat purchases drop?”
- Investigate retention
- Choose a defined screen
Returning-customer chart
Period comparison
If more context is needed: “Which period should we compare?”
Fixed: Overview → Reports → Choose a chartIllustration only. No live integration.
Questions, answered
What is intent-driven UI?
Intent Driven UI adapts the next screen to what someone needs. A product defines the available screens and rules; a model interprets the situation and helps choose what to show next.
What does JEV do in this demo?
JEV interprets ride needs and returns decisions from defined choices, with confidence signals. The app uses those decisions to show designed components for pickup, first-ride reassurance, luggage, or pet-travel guidance. It can also ask for clarification.
How is JEV different from a general-purpose LLM?
Both can interpret intent and return structured choices. General-purpose LLMs also support open-ended language generation and reasoning. JEV specializes in fast, bounded, typed decisions. In this demo, the model selects; it does not write the interface.
Which examples can I try?
The rideshare demo uses live JEV decisions. Choose a situation or type your own, then follow Trip, Preferences, and Review. You can also record audio and review its transcript. E-commerce, onboarding, and insights are illustrations without live integrations.
Can JEV choose the wrong screen?
Yes. A valid choice can still be wrong when the input is unclear, incomplete, or unfamiliar. Confidence shows how sure the model is; it does not guarantee correctness. You can edit your situation or choose a supported need.
Does the demo book a ride or save my recording?
No ride is booked. Preferences and recording state stay in the current session. Audio is sent for transcription, and reviewed text is sent for model processing through OpenRouter. The app does not save recordings or transcripts to disk.
Who created Intent Driven UI?
Madhuri Maram created this design experiment to explore how model decisions can connect a person’s situation to useful interface components.
A design experiment by Madhuri Maram.