Recently, I’ve been thinking about how to design education-specific UX within the confines of a chat-based interface.
Building , my exploratory AI learning-agent app aimed at adult self-learners, has given me a place to test some of these ideas.
In this piece, I’m sharing nine UX details from Sixtus that I’ve found helpful for learning in a chat-based interface, and why I decided to build them.
Learning Path
A general-purpose chatbot has no direction and gives you all the information at once. I wanted Sixtus to be proactive, so I gave it a learning path for the session. It updates as the conversation goes on, the agent proactivly steers based on it, and checkpoints get marked off as you complete them. Seeing the bigger picture and ticking off steps makes you want to finish the next one and keep learning.
Oral Exam
I wanted the oral exam feature to cover everything that could happen in a school oral exam, not only the talking. So I gave the voice agent control of the screen. It can put up an image, play a video or an audio clip, and talk you through it while you answer. You get examined the way you would in an in-person oral exam. Material is put in front of you, and someone asks you to walk through it.
Tutor Settings
I wanted it to be easy to change how the tutor talks. How demanding it is, how formal, and the level it teaches at. So I put those controls in the chat. You can shift the personality to match the subject and keep going.
Lookup
When you're learning, you often encounter words you don't fully know. The meaning matters, but asking about each one pulls you out of the session. I wanted a one-click lookup, so I made terms clickable. You get the context and keep going in the session.
Real-World Tasks
Most of the learning happens outside the app. You read a chapter, watch a lecture, try something. I didn't want the chat to be in competition with that, so Sixtus can hand out a task: go read this, come back when you're done. You confirm it, and the session picks up from there.
Tangent
While you learn, side questions come up. Asking them in the main chat has two costs. The agent's context fills with things that aren't the subject, and the chat gets long enough that the main thread is hard to find. So I added tangents feature, that is like a branched side conversation.
Reminders
Most of what you learn in a session fades by the next one. Spaced repetition fixes that, but it only works if it's in front of you. So the agent sets reminders. You get asked to show what you still retain knowledge from earlier sessions, without having to remember to do it yourself.
Glossary
Over a long session, terms pile up, especially when you come back to an old conversation. So there's a glossary button. One click gives you every term from the session, defined in context, and you're caught up.
Chat Question
After an explanation, you should have to show you got it. A comprehensive assement does that, but it's very time consuming. I wanted something quick along side it. So Sixtus drops a short question of different kinds into the chat after it explains something.
What’s next
Sixtus is still in beta, so these are just some of the ideas I’m exploring, and they’re likely to change as I continue iterating on the project.