09

Daily quests with an AI game master

Own product · PWA, generation under context, gamification

Every morning the app hands over one quest of 5–15 minutes: the AI builds it around the city, the weather, the mood and what the person has already done. Mark it done and the experience lands, the streak grows. It installs on a phone like an app and sends a push.

How it worked before

To-do apps ask the person to decide what to do. That is the step most people stop at: choosing the thing costs more than doing it.

What made it hard

A generated task has to be doable today and here: a quest for a downpour in an unfamiliar city is not the same as one for sunshine two steps from home. And all of it has to run without a server anyone keeps alive by hand.

Engineering decisions

  1. 01

    The context is assembled before the model is called

    City, weather, the marked mood and the history of what was done are composed into a picture of the day — the model receives that, not a bare "invent a task". Quality here is set by what goes in, not by the length of the prompt.

  2. 02

    The quest is prepared on a schedule, not on open

    It is generated ahead of time by cron, so the app opens with a quest ready rather than spinning while someone waits. The most expensive operation happens where nobody is watching it.

  3. 03

    Gamification lives in separate modules

    Experience, levels and streaks are the core; the reward shop and the money bank plug in on top and may not exist at all. The core does not know about them, so switching a module off cannot break progress.

  4. 04

    No server to keep alive

    The schedule, the queue and the pushes live in the platform rather than in a process someone has to keep running. A product for one person should not require administration.

Where it stands

The app opens with a quest ready rather than a spinner. The progress core keeps working with the reward modules switched off.

City and weather
an external data source
Picture of the day
context composed before the call
Generated by cron
overnight, ahead of time
Quest ready
waiting for the morning
Mood and history
what the system already knows
Push in the morning
the app opens with no wait
Experience and streak
core; rewards are a module
The expensive step runs overnight rather than at the moment someone opens the app. The context is assembled before the model is called — it receives a picture of the day rather than a request to invent something.

Got something like this?

Get in touch