Situation X

Homepage, you can choose any scenario to play or generate one game through an AI agent.
After multiple rounds of Q and A, you need to find the full story.
After solving the task, you will be rated of your performance.
We also tried to incorporate TTS and STT apis to make users directly talk with agents.

Inspiration

Our inspiration came from the way real-life problems often reveal only a partial picture at first glance. To uncover the full story, we repeatedly need to ask questions, verify details, and reason carefully. We realized this natural process of iterative discovery could form the foundation of an engaging game—especially one powered by multiple intelligent agents. By combining lateral thinking puzzles with agent-based AI research, we imagined a platform where players collaborate with autonomous agents to uncover hidden narratives, leading to the creation of Situation X.

What it does

Situation X is an interactive, AI-driven puzzle platform where players encounter mysterious scenarios and collaborate with a team of autonomous AI agents. Players can ask questions, request hints, view dynamic visualizations, and even hear scenarios narrated in real time. The agents adapt to player input, maintain contextual memory, and guide reasoning, creating a rich, immersive puzzle-solving experience that blends strategy, logic, and AI interaction.

How we built it

Situation X is powered by a coordinated network of six specialized agents:

Story Generation Agent – creates scenarios based on user-selected difficulty levels.
Game Master Agent – answers yes/no questions using scenario logic.
Hint Agent – provides context-aware, incrementally specific hints without revealing the solution.
Evaluation Agent – assesses whether the player-submitted story matches the original scenario.
Visualization Agent – generates real-time scenario images using a Text-to-Image API.
TTS Agent – narrates scenarios and agent responses, and supports voice input from users.

For the more complex tasks, the first four agents are powered by Sonnet4.5 as the backbone, while simpler, high-frequency tasks like hint generation use Haiku4.5 for efficiency. All agents leverage ClaudeAPI for reasoning, except where task-specific backbones are employed. Scenario data, agent logs, session states, and user ratings are stored in a lightweight SQLite database. We also implemented a scoring system that rewards efficiency and strategic reasoning, factoring in the number of questions asked, hints used, time bonuses, and difficulty multipliers.

The backend is built in Python and Flask, orchestrating agent coordination, session management, and scoring. The frontend is an interactive single-page web interface that allows real-time player-agent interaction. By combining multimodal AI capabilities with structured session management, we created a smooth, immersive gameplay experience that feels both dynamic and intelligent.

Challenges we ran into

One of our main challenges was maintaining consistency across multiple agents and modalities. Each agent must reason coherently while supporting visuals, narration, and hints simultaneously. Balancing agent autonomy and guidance was also difficult: we wanted agents to help without revealing solutions. Additionally, running six agents in parallel created latency and performance constraints, requiring careful optimization. Finally, designing adaptive puzzles that respond to diverse player strategies while remaining fair took multiple iterations.

Accomplishments that we're proud of

We successfully built a fully orchestrated multi-agent AI system that guides players in real time, generating immersive, coherent scenarios that adapt dynamically to user input. Integrating Text-to-Image and TTS APIs brought the agents to life, making the experience multimodal and engaging. The scoring system encourages thoughtful, strategic play, rewarding efficient reasoning and minimal hint use. Our scenarios are replayable, dynamic, and consistently coherent, providing a fresh challenge in every session.

What we learned

We learned that autonomous AI agents can meaningfully enhance human reasoning when they maintain context and memory. Multimodal integration, visuals, narration, and reasoning, significantly improves engagement. Real-time coordination of multiple agents requires careful session management, and adaptive feedback dramatically enhances user experience compared to static puzzles.

What's next for Situation X

We plan to expand agent capabilities to support more sophisticated reasoning, inter-agent collaboration, and adaptive puzzle design. Dynamic scenario generation using generative AI will enable virtually limitless puzzles, while persistent player profiles will allow agents to tailor challenges to individual skill levels. Multiplayer modes will foster collaborative problem-solving with AI, and extending Situation X into mobile and web apps will make the platform accessible anytime, anywhere. By further enhancing the immersive multimodal experience—integrating visuals, narration, and intelligent feedback—Situation X is poised to become a scalable, cutting-edge AI-driven puzzle ecosystem that merges gameplay with advanced AI research.