theunpartybeta

Location: https://github.com/unparty-app/theunpartybeta/
Status: Active Development
Primary Purpose: iOS/macOS native machine learning application with on-device text classification, clustering, and perceptron-based intelligence

Tech Stack

• Framework: SwiftUI
• Language: Swift 5.0
• Platform: iOS 14+, macOS 11+ (Universal - iPhone, iPad)
• ML Framework: Core ML, Create ML, Natural Language
• Graphics: Metal Shaders
• Database: SwiftData (local persistence)
• Key Tools: Xcode, Create ML

Key Features

Machine Learning & Intelligence

• Custom Perceptron Service: On-device binary text classification (action vs non-action)
• Text Classification Model: UNPARTYTextClassifier trained with Create ML
• K-Means Clustering: Enhanced clustering service with multiple distance metrics
• Word Embeddings: Custom word embedding service for semantic analysis
• Adaptive Configuration: Device-aware ML configuration based on memory and capabilities
• Training Management: Comprehensive training data collection and model refinement

Neural Network Visualization

• Neuron Activation View: Interactive visualization of neural network activations
• Custom Neuron Components: Modular neuron implementation with activation functions
• Hidden Layer Architecture: Multi-layer neural network support
• Metal Shaders: Hardware-accelerated background animations and visual effects

User Experience

• Entry System: Rich text entry creation and management
• Search Functionality: Semantic search across user entries with caching
• Quiz System: Interactive quiz interface for training model preferences
• Store Interface: Content storage and retrieval system
• User Profile: Photo management and personalization
• Settings Management: Comprehensive configuration options

Storage & Persistence

• Local Storage Provider: File-based persistence for ML models and user data
• SwiftData Integration: Modern data persistence with @Model and @Query
• Training Data Management: Organized storage for ML training examples
• Model Persistence: Save and load trained ML models

Architecture

Native App (Swift + SwiftUI)
├── Models
│   ├── ClusterModel - Clustering data structures
│   ├── EntryModel - User entry management
│   ├── MoodModel - Sentiment tracking
│   ├── ContentMetrics - Performance analytics
│   └── TrainTextModel - ML training data
├── ViewModels (MVVM Pattern)
│   ├── Entry - Entry lifecycle management
│   ├── Text - Text processing and classification
│   ├── Cluster - Clustering coordination
│   ├── Action - Action detection logic
│   ├── Mood - Mood tracking
│   ├── Search - Search functionality
│   └── Stats - Analytics and metrics
├── Views (SwiftUI)
│   ├── ContentView - Main navigation split view
│   ├── EntryView - Entry creation/editing
│   ├── SearchView - Search interface
│   ├── StoreView - Content store
│   ├── SettingsView - Configuration
│   ├── QuizView - Training quiz interface
│   ├── NeuronActivationView - Neural network visualization
│   ├── TextClassifier - ML classification views
│   └── TrainingStats - Analytics dashboard
├── Services (Business Logic)
│   ├── ActionPerceptronService - Binary text classification
│   ├── KMeansClusteringService - K-Means implementation
│   ├── EnhancedKMeansClusteringService - Advanced clustering
│   ├── ModelTrainerService - ML model training orchestration
│   ├── WordEmbeddingService - Text embeddings
│   ├── TextTokenizationService - Text preprocessing
│   ├── MoodService - Sentiment analysis
│   ├── SimilarityService - Semantic similarity
│   ├── DeviceCapabilityProvider - Hardware capability detection
│   └── EmbeddingConfiguration - Dynamic ML configuration
├── Managers
│   ├── TrainingDataManager - Training example management
│   ├── ModelPersistenceManager - Model save/load
│   └── QuizManager - Quiz state management
├── Storage
│   ├── StorageProvider - Storage abstraction protocol
│   └── LocalStorageProvider - File-based implementation
├── Neuron (Custom Neural Network)
│   ├── Neuron - Base neuron implementation
│   ├── NeuronViewModel - Neuron state management
│   ├── ActivationFunctions - Sigmoid, ReLU, Tanh, etc.
│   ├── Components - Neuron building blocks
│   ├── HiddenLayers - Multi-layer architecture
│   └── Wrappers - Neuron wrapper utilities
└── ML Assets
    ├── UNPARTYTextClassifier.mlproj - Create ML project
    ├── UNPARTYTextClassifier.mlmodel - Trained model
    └── BackgroundShaders.metal - Metal compute shaders

Integration Points

• Core ML: On-device machine learning inference and training
• Natural Language: Text analysis and linguistic feature extraction
• Metal: GPU-accelerated shaders for visual effects
• SwiftData: Modern Apple data persistence framework
• CloudKit: Potential cloud sync (infrastructure ready)

Business Value

ABOUT: Understanding User Patterns

• Perceptron Classification: Automatically categorizes user input as actionable or informational
• Clustering Intelligence: Groups similar entries to reveal patterns in user behavior and thought processes
• Mood Tracking: Provides sentiment insights across user entries
• Training Analytics: Visualizes model performance and learning progress

BUILD: Content Creation & Intelligence

• Entry System: Rich text entry creation with ML-powered classification
• Model Training: Users can train and refine their personal ML models through entries and quizzes
• Search Intelligence: Semantic search enables quick retrieval based on meaning, not just keywords
• Custom Neural Networks: Extensible neuron architecture for future AI capabilities

CONNECT: User Journey & Feedback

• Interactive Quizzes: Engages users in training their personal ML models
• Visual Feedback: Neuron activation view provides transparency into ML decision-making
• Training Stats: Comprehensive analytics dashboard shows model improvement over time
• Store System: Organized content management for sharing and review

Relationship to UNPARTY Ecosystem

Position in Ecosystem

theunpartybeta is a native iOS/macOS application focused on on-device machine learning and intelligent text processing. It represents an experimental branch of the UNPARTY platform, exploring:

• Personal ML model training without cloud dependencies
• On-device privacy-first intelligence
• Custom neural network implementations
• Hardware-accelerated visualizations

Data Flow

Ecosystem Alignment

Distinguishing Features

Unlike theunpartyunppp (CloudKit-based journal app) or theunpartyapp (Next.js web platform), theunpartybeta focuses on:

1. Experimental ML: Testing perceptron and clustering algorithms for eventual integration
2. On-Device Intelligence: Privacy-first ML without cloud dependencies
3. Neural Network Visualization: Educational interface for understanding AI decisions
4. Hardware Acceleration: Metal shaders for performance-critical operations

Development Setup

Prerequisites

• macOS: Ventura (13.0) or later
• Xcode: 14.0 or later
• Swift: 5.0 or later
• Target Devices: iOS 14+, macOS 11+

Getting Started

1. Clone the repository

   bash

   git clone https://github.com/unparty-app/theunpartybeta.git
   cd theunpartybeta

   CopyCopied!

2. Open in Xcode

   bash

   open unpartyappbeta.xcodeproj

   CopyCopied!

3. Configure Signing

   • Select the unpartyappbeta target
   • Update Team to your Apple Developer account
   • Bundle Identifier: UNPARTY.unpartyappbeta

4. Build and Run

   • Select target device (iOS Simulator or Mac)
   • Press ⌘R to build and run

Project Structure

• unpartyappbeta/ - Main application source code
• unpartyappbeta.xcodeproj/ - Xcode project configuration
• UNPARTYTextClassifier.mlproj/ - Create ML training project
• BackgroundShaders.metal - Metal shader source

Machine Learning Details

Text Classification Pipeline

1. Input: User enters text in entry view
2. Preprocessing: TextTokenizationService extracts features
3. Classification: ActionPerceptronService or UNPARTYTextClassifier processes text
4. Result: Binary classification (action/non-action) with confidence score

Clustering Workflow

1. Data Collection: Entries stored via LocalStorageProvider
2. Embedding: WordEmbeddingService generates vector representations
3. Clustering: KMeansClusteringService groups similar entries
4. Visualization: ClusterView displays results

Training Process

1. User Feedback: Quiz responses and manual classifications
2. Data Storage: TrainingDataManager persists examples
3. Model Update: ModelTrainerService retrains perceptron
4. Persistence: ModelPersistenceManager saves updated weights

Device Adaptation

// Automatic configuration based on device capabilities
let config = try EmbeddingConfiguration.from(
    device: DeviceCapabilityProvider.current,
    userPreferences: .default,
    contentMetrics: ContentMetrics.default
)

Key Technologies Explained

Perceptron Service

• Type: Binary linear classifier
• Algorithm: Weighted sum with threshold
• Features: Action keywords, imperative tone detection
• Training: Online learning from user feedback
• Use Case: Categorizing text as actionable vs informational

K-Means Clustering

• Implementation: Enhanced with multiple distance metrics (Euclidean, Cosine, Manhattan)
• Initialization: K-Means++ for better convergence
• Adaptation: Dynamic K selection based on data characteristics
• Use Case: Grouping similar entries for pattern discovery

Metal Shaders

• File: BackgroundShaders.metal
• Purpose: Hardware-accelerated background animations
• Integration: SwiftUI with Metal compute kernels
• Performance: GPU-based rendering for smooth 60fps visuals

Testing & Quality

ML Model Validation

• Training data integrity checks
• Classification accuracy tracking
• Clustering quality metrics (silhouette score)
• Device capability testing

UI Testing Targets

• Entry creation and editing flow
• Search functionality
• Quiz interaction
• Settings management
• Neuron visualization rendering

Future Roadmap

Planned Features

• Export trained models for use in other UNPARTY apps
• Cloud sync option while maintaining privacy (encrypted)
• Transformer-based embeddings for improved semantic understanding
• Multi-class classification beyond binary action detection
• Collaborative filtering for content recommendations
• Integration with theunpartyapp web platform

Research Areas

• Federated learning across user devices
• Differential privacy in model training
• On-device large language model fine-tuning
• Neural architecture search for optimal model size

Contributing

Code Style

• SwiftLint: Enforces Swift style guide (when configured)
• Architecture: MVVM pattern with protocol-oriented design
• Naming: Descriptive names following Swift conventions
• Comments: Document complex ML algorithms and business logic

Pull Request Process

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-ml-feature)
3. Commit changes with descriptive messages
4. Test on both iOS and macOS targets
5. Submit pull request with detailed description

License

Copyright © 2024 UNPARTY LLC
All rights reserved. This is proprietary software for the UNPARTY ecosystem.

Related Documentation

• theunpartyapp: Web platform with content management (Next.js + PostgreSQL)
• theunpartyunppp: CloudKit-based journaling app (Swift + SwiftUI)
• theunpartyrunway: Development automation and intelligence (Python + Bash)
• theunpartycrawler: Analytics intelligence and conversation processing (Python)

Contact & Support

• Developer: UNPARTY LLC
• Bundle ID: UNPARTY.unpartyappbeta
• Version: 1.0
• Platform: iOS 14+, macOS 11+

Status: 🔬 Experimental - Active ML Research & Development
Focus: Privacy-first, on-device machine learning with creator ownership and cost-sensitivity

Metrics

Last Updated: 2025-10-29
Part of the UNPARTY Ecosystem - Measurable user progress through ABOUT → BUILD → CONNECT while protecting creator ownership, privacy, and cost-sensitivity.