🤖 E.D.I. — Enhanced Digital Intelligence

Social Vision Edition v2.0

E.D.I. (pronounced Edie) is a local-first, multi-modal social AI assistant that combines computer vision, speaker recognition, and a persistent relational memory system.

This project is not just a runtime application — it builds a personal AI database over time, storing biometric encodings and long-term knowledge locally on disk.

---

✨ Key Features

• Real-time object detection using YOLOv8 (GPU / CUDA)
• Face recognition without storing raw images
• Voice authentication using speaker embeddings
• Active speaker detection (vision + audio)
• Persistent memory using a Neo4j-style knowledge graph
• Fully local biometric storage (no cloud uploads)

---

🧠 Data Persistence Architecture (IMPORTANT)

E.D.I. continuously builds a structured “brain” inside the data/ directory.
Understanding this layout is essential for backups, migrations, and privacy.

📂 Directory Structure

data/
├── faces/
│   └── face_clusters.json
├── voice/
│   └── voice_clusters.json
├── memory/
│   └── knowledge_graph.json
└── temp/

🧍 data/faces/face_clusters.json

Biometric Face Database

Stores numerical face encodings (no raw images)

Each person has a cluster of vectors captured from:

Front

Left

Right

Up

Down

Tracks quality and confidence scores to prevent learning blurry frames

Used for long-term identity recognition

🎤 data/voice/voice_clusters.json

Speaker Recognition Database

Stores 256-dimensional voice embeddings

Generated during vocal authorization

Allows identity recognition even when the user is off-camera

Used by the VoiceAuth module

🧠 data/memory/knowledge_graph.json

Persistent Relational Memory (Edie’s “Soul”)

Neo4j-style relational graph stored as JSON

Stores:

Facts (e.g., Parth → LIKES → Coffee)

Relationships (e.g., User → WORKS_ON → Project)

Episodic summaries of past interactions

Enables long-term personalization across sessions

⚠️ Deleting this file resets Edie’s personality and memory.

🧪 data/temp/

Temporary Runtime Storage

Used for:

Generated .wav files (Text-to-Speech)

Short-lived intermediate runtime data

Automatically cleared by the system

Safe to delete at any time

🛠️ System Requirements Hardware NVIDIA GPU (recommended for YOLOv8)

CUDA-compatible drivers

Webcam and microphone

Software Python 3.10

Conda (recommended)

NVIDIA drivers + CUDA toolkit

📦 requirements.txt (Version-Pinned) Due to compatibility issues between MediaPipe and NumPy 2.x, strict version pinning is required.

numpy==1.26.4
mediapipe==0.10.11
opencv-python==4.8.0.74
protobuf==3.20.3
ultralytics
google-generativeai
resemblyzer
face-recognition

⚠️ Installing NumPy 2.x will crash MediaPipe.

🚀 Installation & Setup

1️⃣ Clone the Repository

git clone https://github.com/yourusername/EDI.git
cd EDI

2️⃣ Create Conda Environment

conda create -n ai_lab python=3.10
conda activate ai_lab

3️⃣ Install Dependencies

pip install -r requirements.txt

4️⃣ Hardware Check Ensure NVIDIA drivers are installed and CUDA is available:

nvidia-smi

5️⃣ Run E.D.I.

python main.py

🎮 First-Run Onboarding When E.D.I. detects a new person, it initiates identity calibration.

Face Scan

Captures five head orientations

Builds a face encoding cluster

Voice Authorization

Records a short authorization phrase

Generates a 256-dimensional speaker embedding

Knowledge Graph Initialization

Creates a persistent identity node

Links future memories and preferences

🔒 Privacy & Security Notice

This project stores biometric data locally on disk.

IMPORTANT

If you plan to publish your fork or make the repository public:

DO NOT commit the data/ directory

DO NOT commit .env files

Your face data, voice data, and memory graph should remain private.

🛡️ Recommended .gitignore

Virtual environments

.env .venv ai_lab/

Biometric & memory data

data/

Python cache

pycache/ *.pyc

OS files

.DS_Store Thumbs.db 🧩 Project Philosophy E.D.I. is designed as a stateful, embodied AI system, not a stateless chatbot. Its intelligence emerges over time through perception, memory, and interaction.