FaceJudge - Enhanced Version

A PyQt6 + OpenCV desktop application with AI-powered roasting and glazing, powered by Gemini LLM and ElevenLabs TTS.

✨ New Features

1. Profile Scraping & Caching

• Automatically scrapes GitHub and LinkedIn profiles in the background
• Caches all scraped data locally in profiles_cache.json
• Only re-scrapes if profile URLs change (smart caching)
• Supports resume upload (PDF, DOCX, TXT) or inline text input

2. AI-Powered Roasting/Glazing

Press Tab to trigger the main AI call with:

• Live screenshot (Base64-encoded face crop)
• OpenCV face metrics (FWHR, jaw ratio, symmetry, etc.)
• Cached LinkedIn & GitHub profile data
• Resume text
• Detected facial emotion

Three modes available:

• Glaze Mode (👍 up gesture): Hyperbolic compliments
• Hate Mode (👎 down gesture): Savage roasts
• Super Hate Mode (two 👎👎): Maximum unhinged chaos

3. Screen Recording

Press r to toggle screen recording.
Videos are saved to captured-videos/ folder with timestamp.
Red indicator dot appears in top-left when recording.

4. Enhanced UI

• Profile panel with resume text input field
• "SAVING..." → "SAVED ✓" checkmark flow
• Checkmark resets when inputs are edited
• Updated hint text showing all keybinds

5. Audio Output

AI responses are automatically passed to ElevenLabs TTS and played aloud.

🚀 Quick Start

Prerequisites

# All required packages:
pip install PyQt6 opencv-python mediapipe numpy requests elevenlabs python-docx PyPDF2

Setup Environment

Create .env file in project root:

GEMINI_API_KEY=your_key_here
# OR
GOOGLE_API_KEY=your_key_here

ELEVENLABS_API_KEY=your_key_here

Run the App

python app.py

⌨️ Keybinds

Key	Action
Tab	Trigger AI (roast/glaze/super-hate based on mode)
r	Toggle screen recording (saves to captured-videos/)
Space	Capture screenshot of current face
m	Toggle face mesh visualization
Click	Cycle to next detected face
Esc	Quit

📁 File Structure

.
├── app.py                    # Main PyQt6 GUI & video loop
├── ai_core.py               # New: AI orchestration, caching, scraping
├── cvModule.py              # Face analysis & metrics
├── mainAI.py                # Gemini LLM integration (refactored)
├── githubinfo.py            # GitHub profile scraper
├── linkedin_profile.py       # LinkedIn profile scraper
├── profiles_cache.json       # Auto-generated: cached profiles
├── captured-videos/         # Auto-created: screen recording videos
├── captures/                # Face capture screenshots
├── templates/index.html     # (Legacy Flask template)
└── .env                      # API keys (not committed)

🔧 How It Works

Profile Scraping Flow

1. User inputs GitHub username, LinkedIn username, resume (file or text)
2. Click SAVE → background thread starts scraping
3. Data cached locally with unique keys (github::username, linkedin::username, etc.)
4. "SAVED ✓" checkmark appears when done
5. Checkmark resets if user edits any field

AI Generation Flow (Tab Key)

1. Screenshot captured & encoded to Base64
2. Face metrics extracted (emotion, FWHR, symmetry, etc.)
3. Cached profiles looked up from profiles_cache.json
4. Payload constructed with all data
5. Gemini LLM called with appropriate system prompt (glaze/hate/super_hate)
6. Response text passed to ElevenLabs TTS
7. Audio plays automatically
8. Response printed to console

Gesture Recognition

• Thumb Side → Neutral (no action)
• Thumb Up → Glaze Mode (compliments)
• Thumb Down → Hate Mode (roasts)
• Two Thumbs Down → Super Hate Mode (unhinged)

📊 System Prompts

Glaze Mode

Creative, poetic, over-the-top compliments combining facial features with real achievements.

Hate Mode

Savage but clever roasts merging facial features with online footprint.

Super Hate Mode

Maximum unhinged chaos - deranged, creative, hilariously brutal.

🛠️ Architecture Notes

ai_core.py (new core module):

• schedule_background_scrape() - runs scraping in daemon thread
• build_payload() - assembles API request with cached data
• call_ai_and_speak() - calls Gemini + ElevenLabs
• Caching system with file locks for thread safety

mainAI.py refactored:

• Wrapped interactive code in if __name__ == "__main__": block
• Made module importable without prompts

🐛 Troubleshooting

"SAVED ✓ not appearing?"

• Check .env has valid API keys
• See browser console for scraping errors (printed to stdout)

"No audio?"

• Verify ELEVENLABS_API_KEY is set
• Check system volume

"Recording not working?"

• Ensure captured-videos/ folder is writable
• Check OpenCV VideoWriter support on your system

"Import errors?"

• Activate venv: .\.venv\Scripts\Activate.ps1 (Windows) or source .venv/bin/activate (Unix)
• Install all deps: pip install -r requirements.txt (if exists)

📝 Future Enhancements

• Cache expiration (refresh data after N days)
• Confidence scores for scraped data
• Multiple voice selections for TTS
• Custom system prompts via UI
• Video preview before upload
• Analytics dashboard

📄 License

(Add your license here)