NeuroForge Documentation

Overview

NeuroForge is an AI prompt testing and comparison platform that allows users to battle three leading AI models (GPT, Claude, and Gemini) against each other in real-time. The platform evaluates responses across multiple dimensions and determines winners through objective scoring, creating an engaging tournament-style experience for AI interaction.

Core Features

AI Battle System

• Three-Model Comparison: Every prompt is simultaneously sent to GPT, Claude, and Gemini models
• Real-time Responses: All models respond in parallel for fair comparison
• Winner Determination: Automatic scoring determines the best response for each prompt
• Tournament Mode: Analysis feature creates elimination-style competitions

Intelligent Scoring Algorithm

The platform evaluates AI responses across six key dimensions:

• Success (15 pts): Error-free responses
• Speed (15 pts): Response time optimization
• Completeness (20 pts): Thoroughness of answers
• Accuracy (20 pts): Relevance and confidence indicators
• Readability (20 pts): Clarity using Flesch Reading Ease
• Structure (10 pts): Organization and formatting

Analysis & Tournament System

• Click Analyze to have other models critique a response
• Analyzed models get disqualified
• Battle continues until one model remains

Thread Management

• Conversation context across all three models
• Persistent conversations with history
• Automatic thread naming

User Interface

Main Dashboard

• Header: Remaining prompts, settings, logout
• Sidebar: Manage threads
• Battle Results: Winner announcements with scores
• Response Cards: Side-by-side comparison
• Expand Modal: Full-screen response reading

Response Features

• Syntax-highlighted code blocks
• Copy buttons for responses
• Detailed score breakdowns
• Status indicators for pass/fail/disqualified

Model Customization

• Personalities (e.g. Sarcastic GPT, Chess Master Claude, Friendly Teacher Gemini)
• Version selection (GPT-4o, Claude Opus 4, Gemini 2.5, etc.)

Technical Architecture

Frontend (React)

frontend/
├── src/
│   ├── components/   # Forge, Login, ModelBox, ThreadSidebar
│   ├── shared/       # modelMap.js, personalities.js
│   ├── utils/        # personality utilities
│   ├── App.jsx       # Main app
│   └── SettingsContext.jsx

Backend (Node.js/Express)

backend/
├── routes/           # forge.js, admin.js
├── auth/             # verifyToken.js
├── utils/            # analyze.js, threadManager.js
├── shared/           # modelMap.js, personalities.js
├── config/           # limits.js
└── server.js

Full Architecture

    neuroforge/
├── server.js                  # Express app entry
├── routes/
│   ├── forge.js              # Main battle endpoints
│   └── admin.js              # Usage monitoring
├── utils/
│   ├── analyze.js            # Response scoring algorithms
│   ├── threadManager.js      # Conversation management
│   ├── promptLimiter.js      # Usage tracking
│   └── modelNames.js         # Model name formatting
├── auth/
│   └── verifyToken.js        # JWT authentication
├── shared/
│   ├── modelMap.js           # Model identifier mapping
│   └── personalities.js      # AI personality definitions
├── gen-token.js              # Token generator CLI
├── frontend/                 # React frontend
│   ├── src/
│   │   ├── components/
│   │   │   ├── Forge.jsx     # Main dashboard
│   │   │   ├── Login.jsx     # Authentication
│   │   │   ├── ModelBox.jsx  # AI response cards
│   │   │   ├── ThreadSidebar.jsx # Thread navigation
│   │   │   ├── ThreadMessages.jsx # Conversation history
│   │   │   ├── SettingsModal.jsx # Configuration panel
│   │   │   └── Footer.jsx    # Application footer
│   │   ├── shared/
│   │   │   └── modelMap.js   # Frontend model utilities
│   │   ├── App.jsx           # Main application
│   │   └── SettingsContext.jsx # Global settings
│   └── package.json
└── package.json

API Endpoints

• POST /api/forge – Submit prompts
• GET /api/admin/usage – Usage stats
• GET /api/threads – List threads
• POST /api/threads – Create thread
• POST /api/analyze – Analyze response

Authentication & Access Control

• JWT tokens with usage limits
• Prompt-based consumption
• Admin reset controls

Scoring Algorithm Details

1. Parse response
2. Apply metrics
3. Normalize scores
4. Aggregate results
5. Select winner

Metrics include: word count, readability, structure, confidence, response time, and accuracy.

Model Integration

• OpenAI GPT: GPT-4o, GPT-4o Mini, GPT-4.1, GPT-5 series
• Anthropic Claude: Claude Opus 4, Claude Sonnet 4, Claude 3 Haiku
• Google Gemini: Gemini 2.5 Flash, Gemini 2.5 Flash Lite

Data Management

• Thread storage with metadata
• Usage tracking per token
• Admin visibility and resets

Security Features

• JWT token validation + expiration
• Role separation for admins
• Input sanitization + error handling

Development Notes

• Modular architecture
• Plugin-ready personality system
• Configurable scoring
• Parallel API calls for speed

Getting Started

Requirements

• Valid token - Please request one for free
• Modern browser (Chrome,Edge,Opera,Mozilla,etc)
• Internet connection

Basic Usage Flow

1. Login with token
2. Create or continue thread
3. Submit prompt
4. View responses and scores
5. Analyze for tournament elimination

Best Practices

• Use descriptive prompts
• Experiment with personalities
• Monitor prompt usage

This documentation covers NeuroForge’s current version (NeuroForge v1.2b). For technical support, contact the development team.