Generate inside rules
Game length, clue sequence, hint limits and host context constrain the generated experience.
External beta submitted · native iOS
Hunt of Secrets turns a real place into a private AI clue game.
A host photographs real objects, the system builds a coherent hunt, and players solve clues against the same physical space. The difficult part is not generating text; it is keeping story, rules, privacy and camera proof aligned.
StatusBuild 3 valid and App Store eligible; external beta submitted.
UserA host creating a private hunt for a room, home, office or event.
RiskGenerated clues must remain playable against real objects.
Evidence93 native unit tests, one UI flow and 44 backend tests.
Current product proof
The experience connects creation, invitation, play and physical verification.
The host can start with a prepared game or photograph objects for a custom hunt. Players join privately, follow clues and confirm the intended object through the camera flow.
- CreateObject photos become context for story, clues and hints.
- PlayHost and player flows remain separate but share one game state.
- VerifyApple Vision reference-photo matching rejects arbitrary objects.
- ReleaseValid IPA, processed screenshots, privacy, support and deletion routes.
What the work demonstrates
AI behavior needs a product contract.
The model is only one layer. The product still needs deterministic rules, bounded output, account ownership, testable game state and a fallback when confidence is low.
Camera as evidence
The product uses native image capture and reference matching instead of trusting a generic confirmation tap.
Store package, not demo
Authentication, backend functions, legal routes, screenshots, archive validation and beta review move together.
Honest boundary
The product is not public on the App Store yet.
The evidence proves a valid build, internal beta, external beta submission and release package. Final external review, App Privacy completion, real-device refresh and public distribution remain separate release gates.
ProvedNative and backend quality gates plus valid store delivery.
OpenExternal approval, final privacy state and public launch.
Relevant to interactive AI products
Building AI that must stay coherent in the real world?
Send the first user action, the evidence the system can trust and what must happen when the model is uncertain.