🚀 Real-Time AI Privacy Display (REAI)

An intelligent, low-latency, and privacy-focused full-stack desktop application designed to assist software engineers during technical interviews. It captures real-time meeting audio, performs instant semantic retrieval over your local resume and projects database, and generates contextual suggested answers utilizing Google's Gemini models.

---

📸 Project Interface & Layout

The desktop application is built with a light-themed, pencil-sketch aesthetic featuring glassmorphism elements. It provides visual feedback including a dynamic AI confidence gauge, live transcription streams, and an integrated workspace browser with full multi-tab support.

• App Shell & Glassmorphism Dock
• Dynamic Interview Dashboard
• Profile Context Manager
• ReAI Multi-Tab Web Browser
• System Configuration Console

---

🧠 System Architecture

REAI is built on a concurrent asynchronous pipeline that captures, processes, retrieves, and responds in under two seconds.

graph TD
    User([Candidate])
    Audio[🎙️ Live Audio Input / Text Query]
    UI[🖥️ Electron + React UI]
    BE[⚡ FastAPI Backend]
    Whisper[🎙️ Local Whisper STT]
    FAISS[🗂️ Local FAISS Vector Store]
    Gemini[🧠 Google Gemini LLM]
    DB[(🛢️ SQLite / Postgres)]

    User -->|Speak/Type| UI
    UI -->|Audio / Text| BE
    BE -->|Raw Audio Bytes| Whisper
    Whisper -->|Transcribed Question| BE
    BE -->|Semantic Lookup Query| FAISS
    FAISS -->|Retrieved Resume & Projects Chunks| BE
    BE -->|Custom Personal Prompt| Gemini
    Gemini -->|Contextual Response Answer| BE
    BE -->|Suggested Answer + Confidence Score| UI
    BE -->|Log History| DB
    UI -->|Suggested Answer & Live Insights| User

Loading

---

🛠️ Technology Stack

Domain	Core Technology	Highlights
Frontend	React 18, TypeScript, Electron	ContextBridge Sandboxing, Custom Canvas Gauges
Backend	Python 3.11, FastAPI, SQLAlchemy	Async Sessions, Custom Middleware, Alembic
AI STT	faster-whisper (Local execution)	base model size, Direct io.BytesIO processing
Vector Store	FAISS, sentence-transformers	all-MiniLM-L6-v2 embeddings, In-Memory LRU Eviction Cache
Database	PostgreSQL / SQLite	Multi-index schemas, Async Pg Drivers
CI/CD	GitHub Actions	Automated Linting, Typechecks, and Pytest suites

---

🌟 Core Features & Security Controls

• ReAI Multi-Tab Browser: A fully functional embedded web browser built with Electron <webview> tags. Features persistent background tabs, a smart address bar, and dynamic navigation integration.
• Anti-Screen Share (Anti-Spy): Toggleable canvas/window level block. When enabled, the application blackouts its contents and mouse cursor in standard screensharing software (Zoom, Discord, Microsoft Teams).
• Global Privacy Hotkey: Toggle Privacy Mode on/off immediately using the global key combo: Ctrl + Shift + A + S
• Taskbar Stealth Mode: Dynamically hide the application's taskbar icon from running process listings (use Alt + Tab to switch window focus).

---

⚙️ Environment Configurations

Create a .env file in the root directory from the .env.example template:

Variable	Default	Purpose
APP_NAME	Real-Time AI Privacy Display	The display title of the FastAPI application.
DEBUG	false	Enables reloaders and echo database outputs.
DATABASE_URL	sqlite+aiosqlite:///./data/reai.db	Database connection string (Postgres/SQLite).
SECRET_KEY	your_secret_key	Secret used to sign JWT authentication cookies.
COOKIE_SECURE	false	Enforce SSL secure flag on JWT access cookies.
GEMINI_API_KEY	(empty)	Default system-wide Google Gemini API Key.
GEMINI_MODEL	gemini-2.5-flash	Selected model version.
WHISPER_MODEL_SIZE	base	Models: tiny, base, small, medium
WHISPER_DEVICE	cpu	Processing device: cpu or cuda
EMBEDDING_MODEL	all-MiniLM-L6-v2	Embedding model for semantic index.
FAISS_INDEX_PATH	./data/faiss_indices	Target output folder for vector databases.
RATE_LIMIT_PER_MINUTE	60	Limit requests per client ip.

Important: You also need to generate a Firebase Admin SDK private key and place it in the backend/ directory as firebase_service_account.json. This is required for Firebase services to function properly.

---

🚀 Installation & Deployment

1. Clone and enter directory:

   git clone https://github.com/Modi-Krish/Privacy-Display.git
   cd Privacy-Display

2. Setup environment configs:

   cp .env.example .env
   # Add your Google Gemini API key inside .env or set it dynamically in the settings UI.
   # Ensure you place your firebase_service_account.json in the backend directory.

Backend Setup:

1. Initialize virtual environment:

   cd backend
   python -m venv .venv
   .venv\Scripts\activate   # Linux/macOS: source .venv/bin/activate

2. Install dependencies:

   pip install --upgrade pip
   pip install -r requirements.txt

3. Run the migrations and start the server:

   # Create SQLite directories
   mkdir data
   # Start Uvicorn
   python -m uvicorn app.main:app --host 127.0.0.1 --port 8000 --reload

Frontend Setup:

1. Install and start:

   cd frontend
   npm install
   npm run dev:electron

---

📡 Core API Reference

The backend app exposes a comprehensive API (fully documented under /docs when running).

Authentication Domain

• POST /api/auth/register - Create user account and initialize profile context.
• POST /api/auth/login - Authenticate credentials and write HTTP-only cookies.
• POST /api/auth/logout - Clear user session and clean active cookies.
• GET /api/auth/me - Fetch authenticated user details.

Profile Context Domain

• POST /api/resume/upload - Upload PDF resume, extract text, and rebuild vector search space.
• DELETE /api/resume - Clear resume and re-evaluate search space.
• POST /api/projects - Add project details and re-index.
• PUT /api/projects/{id} - Modify project details and re-index.
• DELETE /api/projects/{id} - Remove project details and re-index.
• POST /api/skills - Add new skill and re-index.
• DELETE /api/skills/{id} - Delete skill and re-index.

Live Session Domain

• POST /api/interview/start - Initialize live session logs.
• POST /api/interview/question - Submit transcription or text question, query FAISS database, query Gemini, score confidence, and log.
• POST /api/interview/end - Terminate session and return statistical summaries.
• POST /api/audio/transcribe - Standalone STT transcription route.

---

🤝 Contribution Guidelines

We welcome contributions from developers! Check out CONTRIBUTING.md and CODE_OF_CONDUCT.md for details on:

1. Coding standard styles and ESLint rules.
2. Pull Request branch strategy.
3. Bug reporting process.

---

🛣️ Future Roadmap

• Dynamic WASM Embeddings: Generate sentence embeddings inside the desktop client using ONNX runtime.
• Multi-turn RAG memory: Inject conversation context histories into prompt builders.
• Offline LLM support: Connect to local endpoints (e.g. Ollama, Llama.cpp) for 100% offline air-gapped security.

---

📜 License

Licensed under the MIT License.