Detective Dino
An experimental mystery game powered by Apple Foundation Models on-device AI.
- Year
- 2026
- Role
- Creative Dev
- Stack
- SwiftUIApple Foundation ModelsApple Intelligence
Concepts I've been exposed to
01
Apple Foundation Models
02
Agentic coding
03
Sense of progression
What is Detective Dino?
Detective Dino is an experimental game that I started developing for fun.
It is about a detective dinosaur whose mission is to solve crimes and find the assassin. The details of the crime scene are asked to the user, who can invent them and create a story from his or her imagination:
- The time when the crime was perpetrated;
- The weather;
- The year;
- The season;
- The weapon of the murder;
- Any extra information to personalize even more the story.
After that, the magic happens: the on-device language model of Apple Intelligence creates a crime scene based on the informations inserted by the user and also generates statements from three suspects who claim they're not guilty, each telling their version of events.
One of them, however, is lying, and the player must figure out who by comparing their statements to the story generated by Apple Intelligence. If the player accuses the real culprit, the case is solved and logged in the statistics.
The more cases you solve, the more points you earn to unlock the most prestigious badges (which activate after a set number of solved cases). These badges appear in the app's main menu and can be used to brag to your friends, proving your skill.
Come on, those are the basics of gaming, what's the point of winning if you cannot flex on others? 😏
What did I learn from this experience?
This was my very first attempt in experimenting with Apple Foundation Models, exploring the on-device elaboration. I've never tried it, in any occasion, but I got the idea after being exposed to it during a workshop.
Then it was a time my girlfriend was really focusing on crime stories and I wanted to build a small experience for her about solving cases and it struck me: "I can use Foundation Models to create an interactive story that always changes, avoiding hardcoded stories that are always the same".
And then I tried for the first time in my life agentic coding, discovering how much it is powerful and accurate the help of AI when it has access to the whole context of the codebase. This improved my workflow immensely, but at the same time I reflected on something:
it is so easy that if used without caution, it could dramatically lower the learning opportunities since it acts directly on the code, without the need of the developer to reason upon the syntax.
So to maximize the learning, the solution is to work with it as a companion, a dictionary, a support to my own ideas!
Attitude in the project
Perseverance.
The game is yet far to be perfect and that is the reason it is still on TestFlight, there are a lot of things to adjust and to understand such as:
- what is the best system prompt that I need to give Apple Intelligence to generate a coherent story that is being followed by the three testimonies?
- how do I let it know about the logical contraddictions that only the guilty suspect has to have?
So it's still a work in progress that constantly stimulates my curiosity.