Protype.AI

Developed by Islam Ibrahim, Director of Carrot Studio

Protype.AI is an advanced conversational AI platform integrating cutting-edge technologies such as Graph Neural Networks (GNN), multimodal intelligence, self-reflection, and autonomous learning. Designed by Carrot Studio, it offers robust knowledge management, continuous learning from diverse sources, and natural user interaction through desktop and web interfaces.

---

Table of Contents

• Overview
• Key Features
• Technical Architecture
• System Requirements
• Installation
• Configuration
• Usage
  • Desktop Application
  • Web Interface
  • Dashboard
  • Teaching the System
  • Continuous Learning
• API Endpoints
• Database Structure
• Deployment
• Contributing
• Future Enhancements
• License
• Contact

---

Overview

Protype.AI transcends traditional AI systems by combining symbolic reasoning, advanced memory, and multimodal capabilities. It supports continuous learning from sources like Wikipedia and Google Gemini, while allowing users to contribute custom knowledge. The system is modular, scalable, and deployable as a desktop app (Tkinter) or a web app (Flask + React).

System Diagram (ASCII)

+----------------+ +----------------+ +----------------+ | User Interface |<----->| Learning Agent |<----->| Knowledge Graph| +----------------+ +----------------+ +----------------+ | | | v v v +----------------+ +----------------+ +----------------+ | Multimodal AI |<----->| Self-Reflection|<----->| Advanced Memory| +----------------+ +----------------+ +----------------+

---

Key Features

1. Self-Reflection

   • Critically evaluates inferences and answers using Chain of Thought (CoT).
   • Generates critical thinking questions to deepen understanding.

2. Advanced Memory

   • Uses FAISS and RAG for semantic search and knowledge retrieval.
   • Stores vast amounts of data efficiently.

3. Multimodal Intelligence

   • Processes text, images (via Gemini Vision/CLIP), and audio (text-to-speech/speech-to-text).
   • Enables natural, multi-format interaction.

4. Autonomous Agent

   • Sets and pursues learning goals independently.
   • Expands knowledge base without human intervention.

5. Knowledge Graph

   • Leverages GNNs to infer complex relationships between concepts.
   • Visualizes knowledge connections.

---

Technical Architecture

Core Modules

1. Learning Manager (learning_manager.py)

   • Coordinates continuous learning from external sources.
   • Trains GNN models and manages reinforcement learning.

2. Autonomous Agent (autonomous_agent.py)

   • Manages self-directed learning goals and web research.

3. Symbolic Reasoning (symbolic_reasoning.py)

   • Solves problems using symbolic math (SymPy) and logical deduction.

4. Self-Reflection (self_reflection.py)

   • Validates inferences and answers for accuracy.

5. Multimodal Intelligence (multimodal_intelligence.py)

   • Integrates text, image, and audio processing.

6. Temporal Awareness (temporal_awareness.py)

   • Handles time-sensitive data and expressions.

7. Knowledge Graph (knowledge_graph.py + graph_neural_network.py)

   • Builds and manages concept relationships with NetworkX and GNNs.

8. Search Engine (search_engine.py)

   • Enhances retrieval with Elasticsearch (optional).

9. Web App (web_app.py)

   • Provides Flask-based API and React frontend.

Data Flow (ASCII)

+----------------+ +----------------+ +----------------+ | User Input |--> | NLP (spaCy/ |--> | Knowledge | | (Text/Image/ | | Gemini) | | Retrieval | | Audio) | +----------------+ +----------------+ +----------------+ | | | v v +----------------+ +----------------+ +----------------+ | Reasoning |<-->| Self-Reflection|--> | Response | | (Symbolic/GNN) | +----------------+ | Generation | +----------------+ +----------------+ | | v v +----------------+ +----------------+ | Knowledge |------------------>| Learning | | Storage | | Updates | +----------------+ +----------------+

Key Libraries

• AI: torch, transformers, torch_geometric, google.generativeai, spacy, sympy
• Data: networkx, numpy, elasticsearch, bs4
• Web: flask, celery
• Media: PIL
• Utilities: requests, json, datetime, pytz, dateutil, threading

---

System Requirements

Minimum

• Python 3.7+
• 4GB RAM
• 2GB disk space

Recommended

• Python 3.9+
• 8GB RAM
• 4GB disk space
• Internet connection
• CUDA-enabled GPU (optional)

---

Installation

1. Clone the Repository

   git clone https://github.com/Carrot-Studio/Protype.AI.git
   cd Protype.AI
   

pip install -r requirements.txt

Required

GEMINI_API_KEY="your_gemini_api_key"

Optional

SERPAPI_KEY="your_serpapi_key" ELASTICSEARCH_URL="http://localhost:9200" DATABASE_URL="postgresql://user:password@localhost:5432/protypeai" REDIS_URL="redis://localhost:6379/0" ELEVENLABS_API_KEY="your_elevenlabs_api_key" Dashboard Navigate to http://localhost:5000/dashboard to:

View system stats (knowledge base size, learning status). Manage knowledge (add/edit/delete entries). Control learning processes. Teaching the System Add custom knowledge programmatically:

Copy from learning_manager import learning_manager learning_manager.add_custom_knowledge( question="What is Protype.AI?", answer="An advanced AI system by Carrot Studio.", source="user" ) Or use the dashboard:

Go to "Knowledge Management." Click "Add New Knowledge." Enter question, answer, and source. Save. Continuous Learning Start/stop learning:

Copy from learning_manager import learning_manager learning_manager.start_learning() # Start learning_manager.stop_learning() # Stop Add topics:

learning_manager.add_topic("Quantum Computing") API Endpoints Endpoint Method Purpose Example Payload /api/chat POST Process user message {"message": "What is AI?"} /api/teach POST Add new knowledge {"question": "X", "answer": "Y"} /api/search POST Search knowledge base {"query": "AI concepts"} /api/knowledge GET Retrieve all knowledge N/A /api/text-to-speech POST Convert text to speech {"text": "Hello world"} Database Structure Table: knowledge Column Type Description Notes id INTEGER Primary key Auto-increment question TEXT Question text Unique, indexed answer TEXT Answer text weight REAL Confidence (0-1) source TEXT Source (e.g., "wikipedia") created_at TIMESTAMP Creation time Default: CURRENT_TIMESTAMP created_by TEXT Creator Default: "system" modified_at TIMESTAMP Last modified time Default: CURRENT_TIMESTAMP modified_by TEXT Last modifier Default: "system Developer: Islam Ibrahim