Case: Unsolved
Mingxuan Li
Advisor: Kari Love
This project explores the design of an interactive detective game that challenges players to reconstruct truth by interrogating dynamically responsive AI suspects, drawing from the narrative traditions of classic mystery fiction and adapting them into a web-based immersive experience.
Project Description
An AI-driven interactive mystery where the truth hides behind every lie.
In CASE: UNSOLVED, you are dropped into the middle of a suspicious student apartment murder. The suspects are AI-powered characters—each with secrets, alibis, and the ability to lie, deny, or confess based on your interrogation style.
This isn’t a simple point-and-click game. You’ll:
Explore a 360° immersive crime scene
Assign your investigation points wisely to analyze evidence and unlock new dialogue
Interrogate suspects whose behavior shifts dynamically depending on your questioning
Untangle red herrings, misleading clues, and twisted timelines
Experience a branching event system that rewards deduction—and punishes oversight
Inspired by Ellery Queen’s 1930s “Reader Challenge” and the golden age of detective fiction, CASE: UNSOLVED is a modern take on classic logic puzzles, merged with generative AI and interactive storytelling.
Every playthrough offers a slightly different experience. The case won’t solve itself. And the AI won’t make it easy.
Can you uncover the truth, or will you be deceived?
Technical Details
1. Dynamic Prompt System for AI Suspects
Instead of pre-scripted dialogues, suspects are powered by layered prompt structures that simulate realistic human reactions under pressure. Each suspect follows a behavioral progression—initial denial, evasive redirection, and eventual confession—triggered dynamically by the player's specific lines of questioning. Prompts are optimized to balance narrative coherence with conversational flexibility, ensuring a responsive and unpredictable interrogation experience.
2. AI-Generated 3D Crime Scene Environments
The investigation space is built using AI-generated 360° panoramic imagery, which is converted into cubemap projections for seamless in-browser navigation. This technique allows players to explore a fully immersive crime scene, uncover hidden clues, and assign investigative points to examine areas of interest, enhancing the realism and strategic depth of evidence gathering.
3. Complex Event and Red Herring System
The narrative progression is governed by a branching event system designed to simulate the uncertainty of real investigations. Players who overlook key clues trigger twists and misleading developments, while correct deductions unlock bonus information and narrative shortcuts. This structure reinforces a cause-and-effect relationship between player decisions and story evolution, encouraging deeper analytical engagement.
Together, these systems merge traditional detective storytelling with modern AI technologies, offering players an experience that is both intellectually challenging and narratively rich.