Using an LLM to make a real decision means wading through walls of text from a model that rarely asks the right questions. Answering five questions in one reply is exhausting.
The goal of this hackathon team will be to create an agentic solution that breaks a tangled decision into a stream of one-at-a-time micro-decisions, choosing the right kind of interaction for each one.
Sometimes that’s a swipe left/right yes/no. Sometimes it’s a slider (“how much do you care about X?”). Sometimes a multiple-choice answer. Sometimes a map, a date range, a rank-these-four. The agent picks the primitive based on what it needs to learn next, ordered by information gain so the most discriminating questions come first.
A live panel can show the option space narrowing in real time, so you feel the decision converging. Every answer is undoable. Tap any prior question to reopen it and the downstream path re-plans.
Comparing concrete options is far easier than writing out a slew of answers to a slew of questions. A pleasing UI takes the cognitive load off the user, resulting in more questions answered, and thus a better final decision.