Skip to content

AI Assistant with Invisible Overlay — Commercial product showcase

License

Notifications You must be signed in to change notification settings

CreatmanCEO/ghost-showcase

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 
 
 
 
 
 
 

Repository files navigation

GHOST — AI Assistant with Invisible Overlay

Real-time AI assistance that stays invisible during screen recordings and screenshots

Status Commercial Platform


Overview

GHOST is an intelligent AI assistant designed for professionals who need real-time support during interviews, meetings, coding sessions, and learning. Unlike traditional assistant tools, GHOST features an invisible overlay that won't appear in screen recordings or screenshots, making it perfect for online interviews and presentations.

Key Innovation: Invisible Technology

Built with Windows' WDA_EXCLUDEFROMCAPTURE flag, GHOST's overlay is completely invisible to screen capture software while remaining fully visible to you. This breakthrough makes it the ideal companion for:

  • Technical interviews — Get coding hints without revealing the assistant
  • Online meetings — Access notes and context while presenting
  • Coding sessions — Real-time code suggestions and documentation
  • Learning — Interactive assistance while following tutorials

Features

🎯 5 Specialized Modes

Each mode is optimized for specific use cases with tailored prompts and behaviors:

  1. Interview Mode — Technical interview support with coding hints and algorithm suggestions
  2. Meeting Mode — Context management and talking points for professional meetings
  3. Coding Mode — Real-time code analysis, suggestions, and documentation lookup
  4. Learning Mode — Interactive explanations and concept clarification
  5. General Mode — Flexible assistance for any task

🤖 Multi-Provider LLM Support

Choose the best model for your needs:

  • Claude (Anthropic) — Primary recommendation for best results
  • GPT (OpenAI) — Alternative for specific use cases
  • Ollama — Run models locally for complete privacy
  • LM Studio — Local model hosting with custom configurations

🎤 Real-Time Speech-to-Text

  • Cloud: Deepgram (ultra-low latency streaming)
  • Local: faster-whisper (offline capability)
  • Push-to-talk and continuous modes
  • Multi-language support

📚 RAG Document Support

  • Upload PDFs, DOCX, TXT files
  • Context-aware responses based on your documents
  • Perfect for interview prep with company materials
  • Automatic document chunking and embedding

🔒 Privacy & Security

  • Offline Mode — Works completely offline with local LLMs
  • No Data Storage — Conversations aren't saved by default
  • Encrypted Config — API keys stored securely
  • Open Architecture — Transparent operation

Architecture

┌─────────────────────────────────────────────┐
│         Electron (TypeScript + React)       │
│  ┌─────────────────────────────────────┐   │
│  │   UI Layer (Invisible Overlay)      │   │
│  │   • WDA_EXCLUDEFROMCAPTURE          │   │
│  │   • Transparent + Always-on-top     │   │
│  │   • Draggable, resizable            │   │
│  └─────────────────────────────────────┘   │
│                    │                        │
│                    ▼                        │
│  ┌─────────────────────────────────────┐   │
│  │   IPC Bridge (WebSocket-like)       │   │
│  └─────────────────────────────────────┘   │
└────────────────┬────────────────────────────┘
                 │
                 ▼
┌─────────────────────────────────────────────┐
│          Python Backend (FastAPI)           │
│  ┌─────────────────────────────────────┐   │
│  │   LLM Router                        │   │
│  │   • Claude API                      │   │
│  │   • OpenAI API                      │   │
│  │   • Ollama (local)                  │   │
│  │   • LM Studio (local)               │   │
│  └─────────────────────────────────────┘   │
│  ┌─────────────────────────────────────┐   │
│  │   STT Engine                        │   │
│  │   • Deepgram (cloud)                │   │
│  │   • faster-whisper (local)          │   │
│  └─────────────────────────────────────┘   │
│  ┌─────────────────────────────────────┐   │
│  │   RAG System                        │   │
│  │   • Document processing             │   │
│  │   • Vector embeddings               │   │
│  │   • Context retrieval               │   │
│  └─────────────────────────────────────┘   │
└─────────────────────────────────────────────┘

See docs/ARCHITECTURE.md for detailed technical documentation.


Tech Stack

Frontend

  • Electron — Desktop application framework
  • React — UI components and state management
  • TypeScript — Type-safe development
  • Tailwind CSS — Styling and responsive design

Backend

  • Python 3.11+ — Core logic and integrations
  • FastAPI — High-performance async API
  • WebSocket — Real-time bidirectional communication

AI & ML

  • Anthropic Claude — Primary LLM
  • OpenAI GPT — Alternative LLM
  • Ollama — Local model hosting
  • Deepgram — Real-time STT
  • faster-whisper — Local STT
  • LangChain — RAG implementation

Infrastructure

  • SQLite — Local data storage
  • ChromaDB — Vector database for RAG
  • PyInstaller — Python backend packaging

Use Cases

1. Technical Interviews

Scenario: Software engineering interview with live coding

How GHOST helps:

  • Suggests algorithms and data structures
  • Provides complexity analysis hints
  • Reminds you of syntax and best practices
  • Invisible to interviewers during screen sharing

Mode: Interview Mode


2. Online Meetings

Scenario: Product demo or client presentation

How GHOST helps:

  • Quick access to talking points
  • Real-time information lookup
  • Context management across topics
  • Professional appearance (no visible assistant)

Mode: Meeting Mode


3. Coding Sessions

Scenario: Building a complex feature

How GHOST helps:

  • Documentation lookup
  • Code pattern suggestions
  • Bug investigation assistance
  • Architecture design discussions

Mode: Coding Mode


4. Learning & Research

Scenario: Following a technical tutorial

How GHOST helps:

  • Concept explanations
  • Related resources
  • Troubleshooting help
  • Interactive Q&A

Mode: Learning Mode


Project Status

Development Stage: Active Development
Target Release: Q2 2026
Platform: Windows 10/11 (macOS planned)

What's Working

  • ✅ Invisible overlay technology
  • ✅ Multi-provider LLM integration
  • ✅ Real-time STT (Deepgram + faster-whisper)
  • ✅ 5 specialized modes
  • ✅ Basic RAG implementation

In Progress

  • 🔨 Advanced RAG with multi-document support
  • 🔨 Voice activity detection (VAD)
  • 🔨 Custom hotkey configuration
  • 🔨 Performance optimizations

Planned

  • 📋 macOS support
  • 📋 Linux support
  • 📋 Mobile companion app
  • 📋 Team/Enterprise features

Documentation


Configuration Example

{
  "llm": {
    "provider": "claude",
    "model": "claude-3-5-sonnet-20241022",
    "temperature": 0.7,
    "max_tokens": 2000
  },
  "stt": {
    "provider": "deepgram",
    "language": "en",
    "mode": "push-to-talk"
  },
  "ui": {
    "opacity": 0.95,
    "position": "top-right",
    "theme": "dark"
  },
  "modes": {
    "default": "general",
    "interview": {
      "system_prompt": "You are a technical interview assistant..."
    }
  }
}

See examples/config.example.json for full configuration.


Why Commercial?

GHOST represents significant R&D investment in:

  • Novel UI technology (invisible overlay implementation)
  • Multi-modal integration (LLM + STT + RAG)
  • User experience design for high-stakes scenarios
  • Performance optimization for real-time operation

This showcase repository demonstrates:

  • Technical capabilities and architectural decisions
  • Real-world problem-solving approach
  • Code quality and design thinking
  • Production-ready engineering practices

Note: This is a documentation showcase. The actual product code is proprietary.


About the Creator

Built by CREATMAN — Full-Stack Developer specializing in automation, AI integration, and production-grade tools.

Other Projects:


License

Commercial Product — All rights reserved.

This showcase repository is for demonstration and portfolio purposes. The actual GHOST application and its source code are proprietary.

For business inquiries: creatmanick@gmail.com


Connect


Built with Electron · React · Python · Claude · Deepgram

About

AI Assistant with Invisible Overlay — Commercial product showcase

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors