AutoReply Assistant

Solve a real-world problem: small businesses and solo founders lose time drafting replies to customer emails and inquiries.
This prototype helps you paste an incoming message and get a high‑quality suggested reply, which you can edit and send.

✨ What this project demonstrates

• AI in the development process: uses an LLM to generate the reply drafts (core feature).
• Working prototype: Flask app with HTML UI (no build step).
• Tests & CI: pytest unit tests and a GitHub Actions workflow.
• Containerized: Dockerfile + docker-compose.

🧱 Tech

• Python 3.10+
• Flask
• OpenAI SDK (openai)
• pytest
• dotenv

🏁 Quickstart

git clone <YOUR_FORK_URL> autoreply-assistant
cd autoreply-assistant
python -m venv .venv
source .venv/bin/activate  # Windows: .venv\Scripts\activate
pip install -r requirements.txt

# Set your OpenAI API key
cp .env.example .env
# then edit .env and set OPENAI_API_KEY

# Run the app
flask --app app/app.py --debug run
# open http://127.0.0.1:5000

🐳 Docker

docker build -t autoreply .
docker run -p 5000:5000 --env-file .env autoreply
# open http://127.0.0.1:5000

🧪 Tests

pytest -q

📁 Project structure

app/
  app.py              # Flask application
  ai.py               # LLM interaction wrapper
  templates/
    base.html
    index.html
  static/
    style.css
tests/
  test_app.py
.github/
  workflows/
    ci.yml
scripts/
  sample_inbox.csv    # Example "inbox" file
  batch_reply.py      # CLI: generate replies for CSV rows
requirements.txt
.env.example
Dockerfile
docker-compose.yml
LICENSE
README.md

🔐 Environment

• OPENAI_API_KEY – your API key
• OPENAI_MODEL (optional, default: gpt-4o-mini)

Note: You can swap the model/provider in app/ai.py.

📝 How to use

1. Paste an incoming message.
2. Optionally choose a tone (friendly, concise, apologetic, etc.) and include key facts (pricing, timing, policy).
3. Click Generate Reply.
4. Edit and copy the draft.

🚀 Extending ideas

• Gmail/Outlook integration (IMAP/Graph APIs).
• Multi‑language replies.
• Team templates and style guides.
• Fine-tune with your historical replies.
• Add guardrails (PII scrubbing, toxicity filters).

---

MIT License © 2025