Rashomon of the Dead - Detective AI Mystery Engine

Ghost Witnesses

5 Spectral Entities

Click a ghost to interrogate. Each witnessed the murder through their biased perspective slice.

Witness Perception Memory

Biased Memory

Select a ghost to view their constrained memory slice index...

👻

Lord Ashford (Victim)

Status: Decaying Ethereal Presence

Gemma Local (Simulated)

CONTRADICTION DETECTED: Statement conflicts with Master Fact Ledger #114

Gemma Ledger Constraint Active Voice Synthesis Playing...

Ground Truth Fact Ledger

Structured immutable database of actual murder events vs witness perception slices.

Ledger Immutable

// Loading Fact Ledger JSON...

Active Contradiction Matrix

No contradictions detected yet. Interrogate witnesses to trigger conflicting claims against the Ground Truth Ledger.

Reconcile Testimonies & Pronounce Sentence

Cross-reference all 5 ghost testimonies against the Fact Ledger inconsistencies. Name the murderer, weapon, and true motive to free the haunted manor.

1. The Culprit

Which ghost lied about their whereabouts and motive?

2. Cause of Death

Disregard ghost misperceptions and match ledger toxicological fact.

3. Primary Motive

Synthesize key statements from Cordelia & Bartholomew.

Hacktoberfest Project Pitch & Architecture

Open-Model Offline AI Detective Game Powered by Gemma & ElevenLabs

Gemma LLM Engine

Runs completely local via Ollama / LM Studio or WebLLM. Restricts prompt contexts strictly to individual witness ledger slices to enforce Rashomon effect.

ElevenLabs Voice Sync

Assigns distinct voice IDs and acoustic timbre to each ghost. Supports API key passthrough or offline Web Speech API fallbacks.

Zero Token Cost

Open-source architecture allows infinite offline mystery generation with deterministic ground truth verification.

Local Setup Instructions (Ollama + Gemma 2)

# 1. Install Ollama & pull Gemma 2 model

ollama run gemma:2b

# 2. Configure CORS headers for browser fetch

OLLAMA_ORIGINS="*" ollama serve

# 3. Connect to local endpoint in Rashomon Settings

Endpoint: http://localhost:11434/api/generate

Hacktoberfest Pull Request Ideas

PR #1: Add dynamic audio spectral visualizer for voice playback
PR #2: Support WebLLM in-browser Gemma execution (Wasm/WebGPU)
PR #3: Expand procedural mystery generator with room mapping
PR #4: Custom ElevenLabs voice clone parameter fine-tuning UI