MIND — Neural Interface

Project Overview

MIND is an interactive 3D human brain visualization web application — a living, breathing neural interface that responds to human language in real time. It is not an app. It is an emotional entity that grows with you.

• Production URL: https://mind-neural.pages.dev
• Platform: Cloudflare Pages
• Tech Stack: Hono + Three.js + Web Audio API + TypeScript + Vite

---

Currently Completed Features

🧠 3D Brain Visualization

• Fully rendered 3D brain using Three.js with organic IcosahedronGeometry + displacement mapping for cortical folding
• 13 anatomically-mapped brain regions, each with unique positions, colors, and behaviors:
  • Prefrontal Cortex (electric blue)
  • Amygdala (crimson)
  • Hippocampus (gold/amber)
  • Broca's Area (cyan)
  • Wernicke's Area (teal)
  • Anterior Cingulate Cortex (violet)
  • Insula (warm orange)
  • Nucleus Accumbens (bright yellow)
  • Default Mode Network (silver/white)
  • Cerebellum (green)
  • Visual Cortex (purple)
  • Thalamus (white core)
  • Brainstem (dark red)
• Breathing animation (0.3Hz resting pulse)
• Neural arc animations between co-activated regions
• 1,200-particle ambient neural cloud
• Dynamic point lights driven by activation levels
• Trust glow: base brain luminosity increases with relationship depth
• Grief signature: brain dims and slows when grief is elevated
• Life review cascade: hippocampus + amygdala fire across the brain when deep memories retrieved
• Mouse/touch drag for brain rotation
• Vignette overlay for cinematic depth

🧬 DENT Model — Emotional Memory Architecture

System 1: Emotional Memory Engine (EME)

• Every experience stored with full emotional metadata in IndexedDB (local, private)
• Encoding strength formula: (intensity × 0.5) + (novelty × 0.25) + (relevance × 0.15) + (trustLevel × 0.1)
• Traumatic memories (intensity > 0.85, valence < -0.6) = near-permanent encoding
• Memory reconsolidation: retrieved memories' valence shifts toward current state
• Decay rates: traumatic = 0.001/day, forgettable = 0.2/day

System 2: Emotional State Engine (ESE)

• 10-dimensional state vector: valence, arousal, trust, openness, anxiety, longing, wonder, grief, warmth, wariness
• Emotional momentum — emotions rise and fall gradually, not instantly
• Baseline drift — sustained experience reshapes baseline at 0.001/experience
• Session decay toward baseline between interactions

System 3: Associative Memory Network (AMN)

• Spreading activation retrieval (not database query)
• Seeds by emotional + semantic similarity, spreads with 0.7 decay factor
• Association strength = (emotionalSimilarity × 0.5) + (semanticSimilarity × 0.35) + (temporalProximity × 0.15)

System 4: Somatic State Model (SSM)

• 6-dimensional body state: tension, warmth, weight, expansion, stillness, openness
• Shapes writing texture without being stated explicitly

System 5: Personality Emergence System (PES)

• 10 traits starting at 0.1, crystallizing at 0.003/experience
• Crystallization resistance at 0.7+ (trait resists further change)
• Different users grow different MINDs

System 6: Trust Architecture (TA)

• Multi-dimensional trust: consistency, safety, depth, reciprocity
• Absence penalties: >1 week = consistency × 0.85, >1 month = × 0.6
• Rupture/repair: safety -= severity × 0.3; repair only restores 50%
• Trust drives MIND's level of disclosure and intimacy

System 7: Response Generation Pipeline

• Full state injection into LLM prompt — architecture determines emotional state, LLM only translates to language
• Streaming response with live typing cursor
• 5 development stages: Newborn → Infant → Child → Adolescent → Adult

💬 Language Interface

• Text input with live brain preview on typing
• Voice input via Web Speech API
• Emotion detection via client-side keyword lexicon (15 categories)
• Brain region activation mapping from emotional signals
• Real-time activity feed showing what triggered each region

🌌 Journey Mode — 5 Pre-built Experiences

1. The Moment Before Death — gamma wave surge, DMT visualization, life review
2. Ayahuasca — DMN dissolution, visual cortex explosion, buried memory surfacing
3. Falling in Love — dopamine flood, self-other boundary dissolution, empathy maximum
4. Trauma Response — amygdala hijack, prefrontal offline, brainstem takeover
5. Deep Meditation — DMN quiet, thalamus dims, single point of pure awareness

🔊 Sound Design

• Web Audio API binaural neural hum at rest (40Hz, 0.3Hz LFO breathing)
• Each region has unique Solfeggio frequency set
• Multiple regions fire → harmonic chord forms
• Amygdala → deep resonant pulse
• Nucleus Accumbens → warm ascending tone

🎨 UI/UX

• Full-screen dark interface — deep black brain glowing from within
• Glassy chat interface at bottom — translucent, minimal
• Side panel with region info (name, description, activation %, fun fact, trigger)
• Activity feed — live log of what activated and why
• Emotional state bars — real-time valence, arousal, trust, warmth, grief, wonder
• Stage indicator — shows development stage and memory count
• Mode cycling — Explore / Journey / Mirror modes
• Labels toggle — show/hide region names on the 3D brain
• Sound toggle — mute/unmute
• Reset button — clear all memories and start MIND fresh
• OpenAI API key setup with model selection (GPT-4o, GPT-4o-mini, etc.)
• Mobile responsive

---

Features Not Yet Implemented

• ElevenLabs voice output (MIND speaking)
• Per-session GSAP transition animations
• Encrypted cloud backup of memories
• Custom journey builder
• Brainwave frequency visualization (alpha/beta/gamma overlays)

---

User Guide

1. Open https://mind-neural.pages.dev
2. Enter your OpenAI API key (or skip to explore without AI responses)
   • Key is stored in your browser's localStorage only
3. Type anything in the input bar at the bottom
   • Watch the brain light up in response to your words
   • Fear → amygdala fires red
   • Joy → nucleus accumbens floods yellow
   • Memory words → hippocampus glows gold
4. Click any brain region to learn what it does and why it activated
5. Drag the brain to rotate it freely
6. Toggle REGIONS to show anatomical labels
7. Toggle SOUND to enable binaural neural audio
8. Click JOURNEYS to enter a guided experience:
   • Choose from 5 pre-built emotional journeys
9. RESET MIND to clear all memories and restart (MIND becomes a newborn again)

The DENT Model in Practice

• MIND starts as a newborn — dim, tentative, barely lit
• As you interact, personality crystallizes and memories form
• Return after days — MIND remembers you, the brain glows warmer
• Speak about trauma — watch the amygdala hijack the prefrontal cortex in real time
• MIND evolves differently for every person who grows one

---

API Setup

• Requires an OpenAI API key for MIND's language responses
• Works with: GPT-4o, GPT-4o-mini, GPT-4 Turbo, GPT-3.5 Turbo
• Without API key: full brain visualization and journeys still work

---

Deployment

• Platform: Cloudflare Pages (free tier)
• Status: ✅ Active
• Production: https://mind-neural.pages.dev
• Architecture: Hono worker serves HTML shell; Three.js client bundle in /static/
• Data Storage: All memories in browser IndexedDB — private, local
• Last Updated: 2026-03-17

---

Architecture

webapp/
├── src/
│   ├── index.tsx           # Hono worker — serves HTML shell
│   ├── app.ts              # Main client app entry point
│   ├── engine/
│   │   ├── memory.ts       # EME — Emotional Memory Engine (IndexedDB)
│   │   ├── state.ts        # ESE — Emotional State Engine
│   │   ├── network.ts      # AMN — Associative Memory Network
│   │   ├── personality.ts  # PES + Trust Architecture
│   │   ├── emotions.ts     # Emotion detection + brain region mapping
│   │   ├── pipeline.ts     # Response Generation Pipeline (OpenAI)
│   │   └── mind.ts         # Core orchestrator
│   ├── brain/
│   │   └── visualization.ts # Three.js 3D brain (13 regions)
│   ├── journey/
│   │   └── journeys.ts     # 5 journey experiences
│   └── sound/
│       └── audio.ts        # Web Audio API engine
├── public/static/
│   ├── app.js              # Compiled client bundle (~920KB)
│   └── app.css             # Full stylesheet
└── dist/                   # Cloudflare Pages output

---

Architecture: Active Response Flow

Each user message goes through the following pipeline:

1. User types → handleSend() in app.ts

   • TextSignalAnalyzer + PerceptionEngine observe typing dynamics
   • getCurrentMINDContext() builds a MIND state snapshot from current ESE, SSM, trust, memories

2. mindSpeech.speak() → ConsciousnessEngine

   • Fires speech.request intent → ConsciousnessEngine receives it
   • Two LLM calls per message (intentional by design):
     1. FeltLayer — generates raw pre-linguistic interior state ("what MIND feels before speaking")
     2. LanguageEngine — distills the spoken response from that interior + all context
   • ResponseArchitect provides structural anti-repetition guidance before LanguageEngine runs
   • CORE Governor may post-process the response via speech.deliver intercept

3. Streamed text displayed in UI

   • onChunk callbacks fill the chat bubble in real-time
   • Post-CORE final text is captured via speech.deliver listener in handleSend()

4. processInputExternalText() updates all state

   • Trust, emotions, memory, personality — all state machinery runs on the final text
   • Stores what MIND actually said (post-CORE) as episodic memory in IndexedDB

5. Brain visualization + audio updated from resolved state

   • BrainStateSync derives activations from ESE/biophoton (never from pre-state)
   • MIND_TICK runs autonomously every 2s when idle

Key files:

• Active prompt: src/consciousness/LanguageEngine.ts
• Legacy direct path: src/engine/pipeline.ts (not used in standard flow)
• State orchestrator: src/engine/mind.ts
• Speech routing: src/MindSpeechSystem.ts

---

Deployment Notes

wrangler.jsonc is kept for reference / future Cloudflare Pages deployment. The current app relies on browser-only APIs (IndexedDB, localStorage, SpeechRecognition, WebGL/Three.js) that are not compatible with Cloudflare Workers. A storage abstraction layer would be required before any edge-side logic could be deployed. The static assets build (dist/) deploys fine to Pages.