A travel companion that speaks first.

Our entry to the Pegasus x Berkeley Haas AI Travel Companion Hackathon. A working mock of the Pegasus app where an agent notices, decides and acts, without ever becoming a chatbot.

Four agents, one face Three mates, one trip to Cappadocia Every number computed, never invented

How we got from a brief to a companion

Five decisions, in the order we made them. Each one cut something so the next could be sharper.

  1. We read what the brief rules out

    The hackathon asks for an AI travel companion inside the Pegasus ecosystem. It rules out a basic chatbot feature list and a standalone assistant disconnected from Pegasus. That became our test for every idea.

  2. We made the companion act on its own

    There is exactly one input, a sentence, and it is answered with a ticket, never with text. No transcript, no reply bubble. Everywhere else the companion decides when to speak.

  3. We storyboarded one journey

    Three mates go to Cappadocia. Jess books first, Archie gets a nudge, Will is the friend without the app, and Dad follows along. Thirteen scenes show every agent at work, in order.

  4. We built it as a real app

    A mock of the Pegasus mobile app, screen for screen, running in the browser. Every figure the companion quotes is arithmetic in code, handed to the screen as a finished sentence, and a test checks that every screen agrees.

  5. We taught it when to stay quiet

    Staying quiet is a feature. Three gates each silence the companion on their own, so a change that makes it speak more is treated as a regression.