Inbox Intelligence — Email Classification + Reply Assistant

A compact, practical system that automatically classifies incoming emails and suggests short professional replies. Built for experimentation and lightweight deployment: FastAPI backend with an ML model, and a minimal React frontend with a polished dark UI.

---

Features

• Fast hybrid classification (keyword heuristics + LinearSVC ML fallback)
• Three labels: high_priority, normal, spam
• Reply suggestion endpoint to generate one-sentence professional replies
• Minimal, responsive React UI with dark mode and smooth animations
• Docker + docker-compose for full-stack local deployment
• Nginx reverse proxy configuration for production-ready static serving and API proxying

---

Architecture

                ┌──────────────────────┐
                │      Browser UI      │
                │  (React, Axios)      │
                └──────────┬───────────┘
                           │
                  (requests to /predict, /suggest_reply)
                           │
             ┌─────────────▼─────────────┐
             │         Nginx proxy       │
             │  (serves frontend build,  │
             │   proxies /api -> backend)│
             └───────┬─────────┬─────────┘
                     │         │
           (80/static)     (8000/api)
                     │         │
        ┌────────────▼─┐   ┌───▼────────┐
        │ Frontend     │   │ Backend     │
        │ (Nginx serve │   │ (FastAPI)   │
        │  static build)│  │  endpoints  │
        └──────────────┘   └─────────────┘
                           │
                           ▼
                 ┌────────────────────┐
                 │  Model artifacts    │
                 │ model.pkl, vectorizer.pkl │
                 └────────────────────┘

---

Tech Stack

• Frontend: React (Hooks), Axios, plain CSS (dark theme)
• Backend: FastAPI, Uvicorn, Pydantic
• ML: scikit-learn (LinearSVC + TfidfVectorizer)
• Deployment: Docker, docker-compose, Nginx

---

Installation (developer/local)

Prerequisites:

• Node.js (v16+ / v18 recommended)
• Python 3.11+
• Docker & Docker Compose (optional for containerized run)

1. Clone repository

git clone <your-repo-url> inbox-intelligence
cd "inbox-intelligence"

2. Backend (local dev)

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

pip install -r requirements.txt
# Run development server
uvicorn app:app --reload --port 8000

Open http://localhost:8000/docs to view the interactive API docs.

3. Frontend (local dev)

cd ../frontend
npm install
npm start

Open http://localhost:3000 (development server).

4. Full stack with Docker Compose

From repository root:

# Build and start everything
docker-compose up --build

# Open UI via the reverse proxy (Nginx):
# http://localhost (port 80)

# Stop
docker-compose down

Notes:

• The frontend calls the backend at http://localhost:8000 in development. When served through Nginx, the proxy routes API calls to the backend service.

---

API Quick Example

POST /predict

Request:

{
  "text": "Your invoice #12345 is overdue. Please remit payment ASAP."
}

Response:

{
  "label": "high_priority",
  "confidence": 1.0
}

POST /suggest_reply

Request:

{
  "label": "high_priority"
}

Response:

{
  "suggested_reply": "Thank you for bringing this to my attention. I will prioritize this and get back to you shortly."
}

---

Future Improvements

• Add probability calibration for meaningful confidence values
• Support email subject, sender parsing and richer metadata
• Add authentication and per-user settings
• Batch processing & saved histories
• Improve ML model (more data, transformer-based classifier)
• A/B testing and model versioning
• Light/dark theme toggle and keyboard shortcuts

---

Enjoy — open http://localhost:3000 and try pasting an email into the app.