Threadville PD

Investigate procedural murders. Collect forensic evidence, interrogate suspects, and solve cases using a dynamic deduction board.

Inspiration

We were inspired by detective games that blend investigative puzzle-solving with authentic forensic gameplay. Traditional procedural generation often feels disconnected from storytelling, so we set out to create a system where AI generates not just random elements, but interconnected narratives—where every piece of evidence has meaning, suspects have believable motives, and solutions emerge from logical forensic connections. The goal: make every case feel like a real case you can actually solve.

What It Does

Threadville PD is a web-based detective investigation game where players:

• Investigate procedurally generated cases - Each case is uniquely crafted by AI with distinct suspects, motives, and evidence
• Collect forensic evidence - Examine photos, documents, forensic samples, and digital artifacts that form the puzzle
• Interrogate suspects - Build profiles and identify inconsistencies in testimonies
• Deduce connections - Place evidence on an interactive investigation board and link connections to form theories
• Solve mysteries - The game validates your deductions in real-time and confirms when you've solved the case

Features include:

• AI-generated narratives powered by Google Gemini (Pro model for logic, Flash for efficiency)
• Generated Game visuals for suspects, crime scenes, and evidence
• Intelligent forensic validation - DNA, fingerprints, and blood types are tracked and must match culprits
• Red herrings and false leads - Not all evidence points to the guilty party
• Dynamic document rendering - Autopsy reports, bank statements, search warrants, and more, generated with contextual data
• Game progression tracking - Resume investigations and track your solving statistics

How We Built It

Tech Stack:

• Frontend: React + TypeScript, Three.js for 3D scene rendering
• Backend: Supabase for database and storage
• AI: Google Gemini 3 API for case generation, image generation, and content enrichment
• Styling: Tailwind CSS + custom skeuomorphic UI components
• Build: Vite + Vitest

Architecture:

1. Case Generation Pipeline

   • AI generates a complete case blueprint with suspects, evidence, and solution
   • Intelligent forensic linking ensures DNA/fingerprints create a solvable logic graph
   • Document enrichment generates contextual "boring list" documents (bank statements, train schedules)
   • Asset image generation creates game visuals

2. Deduction System

   • Automatic rule generation from case data (what connections are correct)
   • Real-time validation of player links against case logic
   • Win condition detection when critical deductions are complete
   • Hint system suggests next logical connections

3. Evidence Management

   • Structured metadata for each piece of evidence
   • Template-based document rendering (autopsy reports, search warrants, etc.)
   • Forensic data validation (DNA sequences, fingerprint patterns, blood types)
   • Support for locked devices with decryption puzzles and audio transcripts

Challenges We Ran Into

1. Procedural Coherence - Initial case generation created disconnected evidence. Solution: Implemented intelligent forensic linking that ensures culprits have matching DNA/fingerprints in evidence, with red herrings for other suspects.

2. Schema Complexity - Gemini's schema validation struggled with deeply nested document templates. Solution: Split document generation into separate calls with targeted schemas instead of sending all metadata at once.

3. Forensic Data Generation - Distinguishing between culprit evidence and red herrings required suspect data visibility during generation. Solution: Embedded suspect forensic profiles early and used them to intelligently assign evidence during case hydration.

4. UI/UX for Investigation - Making the investigation board intuitive for creating and validating connections. Solution: Built a custom React component with drag-and-drop linking and real-time visual feedback.

5. Image Generation Quality - Ensuring assets generated without UI elements or artifacts. Solution: Developed detailed negative prompts and specific style prompts for consistency.

Accomplishments We're Proud Of

✅ Complete procedural case generation - Every aspect of a case (suspects, evidence, documents, solutions) is AI-generated and contextually coherent

✅ Intelligent deduction system - Cases create solvable logic puzzles with real forensic constraints (DNA, fingerprints, blood type matching)

✅ Production-ready architecture - deduction logic, comprehensive validation, and error handling

✅ Multi-layer evidence system - Support for documents, photos, forensic samples, and digital artifacts (audio, locked devices)

✅ Visual assets - Stylized images of suspects and crime scenes that feel game-ready

✅ Scalable design - Modular components and services allow for easy expansion (new document types, forensic categories, game modes)

✅ Real-time feedback - Players know immediately whether their deductions are correct, with clear messaging on what's missing

What We Learned

• Procedural generation requires constraint satisfaction - Random content isn't enough; players need a solvable puzzle with discoverable logic
• Multi-model AI strategies matter - Using Gemini Pro for complex logic and Flash for high-volume tasks balances quality and cost
• Forensic authenticity drives engagement - Players care about real-world investigative mechanics (DNA matching, fingerprint analysis)
• Schema design is critical - Overly complex JSON structures break AI output; smaller, focused schemas work better
• Red herrings are essential - A case without misleading evidence feels too easy; believable false leads create tension

What's Next for Threadville PD

• Mobile optimization - Adapt the investigation board for tablet and mobile experiences
• Multiplayer deduction - Cooperative investigations where players compare theories and share discoveries
• Case difficulty scaling - From "Rookie" (obvious culprits) to "Legendary" (complex misdirection)
• Case library - Let players save favorite cases, replay, and compare solving times
• Achievement system - Track forensic accuracy, speed, and number of cases solved
• DLC content packs - Themed cases (corporate espionage, organized crime, etc.)
• Community cases - Allow players to generate and share cases they've designed
• Crime Scenes - Allow players to move from the desk to the actual crime scenes to collect more evidence.

---

Run Locally

Prerequisites: Node.js

1. Install dependencies: npm install
2. Set the GEMINI_API_KEY in .env.local to your Gemini API key
3. Run the app: npm run dev

View your app online at https://threadvillepd.vercel.app