Lazy Mail

Overview

Lazy Mail is an automated agent that filters and summarizes important emails from Microsoft Outlook using LLMs and delivers a daily briefing via Telegram.

The system addresses inbox overload by reducing noise from newsletters, spam, and phishing attempts, ensuring that relevant messages (such as professional opportunities) are not missed.

Problem

Over the years, my primary email account accumulated:

• Old newsletters
• Recurring promotional emails
• Phishing attempts
• Inconsistent spam filtering (including false positives)

The default spam filter is not always reliable — I have missed important communications due to emails being incorrectly classified.

Daily manual triage became time-consuming and inefficient.

Solution

I developed an automated workflow that:

1. Runs every morning
2. Calls the Microsoft Graph API to collect emails from the previous day
3. Sends the retrieved data to an LLM to perform:
   • Relevance classification
   • Promotional content filtering
   • Summary generation
4. Delivers the output as a newsletter-style message via a Telegram bot

The result is a concise morning briefing containing only the relevant emails from the previous day.

Architecture

• Daily trigger (cron)
• Request to Microsoft Outlook via Graph API
• Processing and filtering through an LLM
• Automated delivery via Telegram bot

Simplified flow:

Technologies Used

• n8n (workflow orchestration)
• Microsoft Graph API
• LLM (classification and summarization)
• Telegram Bot API
• Self-hosted environment (Home Lab)

Technical Decisions

• Used n8n to accelerate prototyping and simplify visual workflow maintenance.
• Leveraged LLM-based processing instead of rule-based filters to allow dynamic adaptation to different email patterns.
• Chose Telegram as the delivery interface for its speed, accessibility, and seamless integration into daily routines.

Technical Challenges

• Designing filters that minimize false negatives without discarding relevant emails.
• Controlling LLM operational costs based on email volume.
• Fine-tuning prompts to ensure concise and objective outputs.

Results

• Eliminated daily manual email triage.
• Significantly reduced the risk of missing important messages.
• Low operational cost (a few cents per day).
• Stable system running daily in a self-hosted environment.

Key Learnings

• Practical integration with the Microsoft Graph API.
• Prompt engineering for classification and summarization tasks.
• Designing automation systems focused on real productivity gains.