Detective Academy

• 

  Project Logo

Inspiration

We were inspired by the challenge of turning AI reasoning into something tangible and interactive. While large language models are often experienced through chat interfaces, we wanted to showcase Gemini 3’s reasoning, multimodal understanding, and structured output capabilities in a more immersive way.

Detective Academy was born from the idea of transforming investigative logic into a playable experience thereby allowing users to analyze evidence, evaluate suspects, and form hypotheses, all while seeing Gemini 3 actively reason through complex, interconnected information.

The goal was to build something that demonstrates Gemini 3 not just as a conversational model, but as a true reasoning engine.

What it does

Detective Academy is an AI-powered investigative reasoning experience.

Players are presented with a crime scene, evidence, and suspect profiles. They analyze clues, form hypotheses, and submit their conclusions. Gemini 3 then evaluates the player’s logic using structured reasoning, returning:

• A match score
• Step-by-step reasoning traces
• Probabilistic suspect likelihoods
• Evidence-based feedback
• Investigative guidance

Each case is powered by Gemini 3’s ability to synthesize narrative context, forensic clues, and user input into coherent multi-step inferences.

The application also demonstrates Gemini’s multimodal capabilities by generating crime scene visuals and reasoning across both textual and visual evidence.

Detective Academy turns Gemini 3 into an active detective partner rather than a passive chatbot.

How we built it

Detective Academy is built with:

• Next.js / React for the frontend UI
• API routes for server-side orchestration
• Google Gemini 3 (Flash) for reasoning, structured outputs, hypothesis evaluation, and guidance
• Gemini 3 multimodal image generation for reconstructing crime scenes
• Vercel for deployment

Architecture overview:

Frontend → API Routes → Gemini 3 → Structured Reasoning Engine → UI

Key Gemini integrations include:

• Structured prompts for investigative evaluation
• Multi-step reasoning pipelines
• Probabilistic suspect scoring
• Multimodal image generation for crime scenes
• Dynamic guidance based on current evidence state

Gemini 3 is central to all gameplay logic, from evaluating player hypotheses to synthesizing forensic contradictions and generating feedback.

Challenges we ran into

One major challenge was designing prompts that consistently produced structured, reliable reasoning across complex narrative inputs.

Balancing creativity with determinism required careful prompt engineering to ensure Gemini could:

• Track multiple suspects
• Compare conflicting evidence
• Maintain session context
• Return machine-readable outputs

Another challenge was exposing AI reasoning transparently in the UI while keeping the experience approachable for players.

We also worked through integrating multimodal generation smoothly into the gameplay loop, ensuring crime scene visuals aligned with narrative details.

Accomplishments that we're proud of

• Building a full interactive experience driven entirely by Gemini 3 reasoning
• Implementing structured logic traces and probabilistic suspect scoring
• Making Gemini’s reasoning visible and understandable to users
• Successfully integrating multimodal image generation into gameplay
• Deploying a live, publicly accessible demo with open-source code

Most importantly, we created a project that demonstrates Gemini 3 as an active reasoning system, not just a text generator.

What we learned

We learned that Gemini 3 excels at multi-step reasoning when given clear structure and context.

We also discovered how powerful it can be to combine narrative design with AI inference — turning abstract model capabilities into concrete user experiences.

Building Detective Academy taught us how to:

• Design AI-first applications
• Create explainable AI outputs
• Orchestrate structured LLM pipelines
• Blend multimodal generation with interactive systems

It reinforced that transparency and explainability are critical for meaningful AI experiences.

What's next for Detective Academy

Next, we plan to:

• Add multiple dynamically generated cases
• Expand Gemini-guided investigative coaching
• Introduce adaptive difficulty based on player performance
• Enhance visual storytelling with richer multimodal scenes
• Explore educational versions for teaching logic and critical thinking

Long-term, Detective Academy could evolve into a platform for demonstrating AI-assisted reasoning in education, simulations, and training environments.

Built With

• api-routes
• gemini-integration
• gemini3
• geminiflash
• next.js-+-react-frontend
• vercel