Overview

This project is an example ai agent for property managers. The app:

• Connects to an email client
• Triages unread messages
• Generates draft replies using an LLM (OpenAI GPT-4.1)
• Triggers relevant workflows (such as creating action items)

Features

• Connects to a Gmail inbox using IMAP (via imaplib)
• Fetches unread or recent messages
• Loads relevant tenant information (mocked for demo purposes)
• Generates draft replies using OpenAI's GPT-4.1 LLM
• Extracts and processes action items (e.g., maintenance requests) as JSON
• Sends email replies to tenants

Setup & Run Instructions

1. Clone the repository
2. Install dependencies:

   pip install openai python-dotenv backoff

3. Set up your .env file with the following variables:

   EMAIL_ADDRESS=[email protected]
   EMAIL_PASSWORD=your_gmail_password_or_app_password
   IMAP_SERVER=imap.gmail.com
   OPENAI_API_KEY=your_openai_api_key

   Note: For Gmail, you may need to use an App Password if 2FA is enabled.

4. Run the app:

   python main.py

Project Structure

• main.py — Entry point, runs the end-to-end cycle
• email_client.py — Handles IMAP connection, email fetching, and sending
• email_handling.py — Loads tenant info, generates LLM replies, processes workflows
• post_processing.py — Extracts draft replies and action items from LLM output

Things to do in the future

• Use actual objects to represent tenants, actions, etc.
• Add persistence
• The classics: add error handling, logging, etc.
• Support HTML email parsing and attachments
• Integrate with a workflow system to create action items

Assumptions

• All relevant tenant data is available or can be mocked
• The LLM can generate useful replies and action items from the email content
• The app is run with access to the relevant inbox

AI Notes

• LLM Used: OpenAI GPT-4.1 via the openai Python SDK
• Prompting: The LLM is prompted to extract intent, generate a draft reply, and suggest an action item in JSON
• Post-processing: Regex is used to extract the draft reply and action item from the LLM output
• Development: I used cusor.

---