Back to Templates
<h2>📧 Analyze, classify, and summarize emails using RAG (automatic taxonomy learning)</h2>
<p>
This workflow automatically reads incoming Gmail messages, analyzes them using AI with a retrieval-augmented
classification system (RAG), organizes emails into structured categories, stores results in Google Sheets,
and even generates an audio summary sent directly to Telegram.
</p>
<hr>
<h2>⚙️ How It Works</h2>
<p>
The workflow creates an intelligent email processing pipeline combining Gmail, OpenAI, vector embeddings,
and Google Sheets to continuously learn and improve email categorization accuracy.
</p>
<ol>
<li>
<strong>Email Trigger</strong><br>
The workflow starts automatically whenever a new email arrives in Gmail using the Gmail Trigger node.
</li>
<li>
<strong>Email Extraction & Cleaning</strong><br>
The full email content, sender information, subject, and timestamp are retrieved and normalized.
HTML formatting is removed so the AI receives clean text data.
</li>
<li>
<strong>Knowledge Base Loading (RAG)</strong><br>
Email tagging examples stored in a Google Sheets document are loaded and converted into embeddings.
These examples act as the official taxonomy for categories and subcategories.
</li>
<li>
<strong>Vector Store Creation</strong><br>
The workflow builds an in-memory vector database containing historical tagging samples.
This allows semantic similarity search during classification.
</li>
<li>
<strong>AI Analysis Agent</strong><br>
An OpenAI-powered AI Agent performs multiple tasks in a single step:
<ul>
<li>Summarizes the email (short paragraph)</li>
<li>Creates a one-line short message</li>
<li>Classifies the email using retrieved vector examples</li>
<li>Extracts keywords</li>
<li>Assigns confidence score</li>
<li>Detects whether a new category or subcategory was created</li>
</ul>
The agent strictly prioritizes retrieved taxonomy examples to maintain consistent labeling.
</li>
<li>
<strong>Structured Data Storage</strong><br>
Processed email data — including summary, category, keywords, and metadata — is appended to a Google Sheets log,
creating a searchable email intelligence database.
</li>
<li>
<strong>Self-Learning Taxonomy</strong><br>
If the AI determines a new category or subcategory is required:
<ul>
<li>The new label is automatically added to the tagging samples sheet</li>
<li>An admin notification is sent via Telegram</li>
</ul>
This allows the system to evolve over time.
</li>
<li>
<strong>Audio Summary Generation</strong><br>
The summarized email text is converted into speech using OpenAI audio generation.
</li>
<li>
<strong>Telegram Notification</strong><br>
The generated audio summary is automatically delivered to Telegram, allowing quick email review without reading.
</li>
</ol>
<hr>
<h2>🧩 Features</h2>
<ul>
<li>✅ Fully automated Gmail monitoring</li>
<li>✅ Retrieval-Augmented Generation (RAG) email classification</li>
<li>✅ Consistent taxonomy enforcement via vector similarity</li>
<li>✅ Automatic keyword extraction & confidence scoring</li>
<li>✅ Self-updating category knowledge base</li>
<li>✅ Google Sheets email database</li>
<li>✅ AI-generated voice summaries</li>
<li>✅ Telegram admin notifications</li>
<li>✅ Continuously improving classification accuracy</li>
</ul>
<hr>
<h2>🚀 Setup Steps</h2>
<h3>1. Required Accounts</h3>
<ul>
<li>Google account (Gmail + Google Sheets)</li>
<li>OpenAI API account</li>
<li>Telegram Bot</li>
<li>n8n instance (Cloud or Self-hosted)</li>
</ul>
<h3>2. Configure Credentials in n8n</h3>
<p>Create the following credentials inside n8n:</p>
<ul>
<li><strong>Gmail OAuth2</strong> – used by Gmail Trigger and message retrieval</li>
<li><strong>Google Sheets OAuth2</strong> – for email storage and taxonomy dataset</li>
<li><strong>OpenAI API</strong> – for embeddings, AI analysis, and audio generation</li>
<li><strong>Telegram API</strong> – for sending notifications and audio summaries</li>
</ul>
<h3>3. Prepare Google Sheets</h3>
<p><strong>Sheet 1 — Email Log</strong></p>
<ul>
<li>date</li>
<li>email</li>
<li>name</li>
<li>subject</li>
<li>summarized text</li>
<li>category</li>
<li>subcategory</li>
<li>keywords</li>
</ul>
<p><strong>Sheet 2 — Tagging Samples (Knowledge Base)</strong></p>
<ul>
<li>id</li>
<li>subject</li>
<li>email_text</li>
<li>category</li>
<li>subcategory</li>
<li>keywords</li>
</ul>
<p>
This sheet acts as the classification training reference used by the vector store.
</p>
<h3>4. Configure Telegram</h3>
<ul>
<li>Create a Telegram bot using BotFather</li>
<li>Copy the bot token into n8n credentials</li>
<li>Replace the <code>chatId</code> value with your Telegram user or group ID</li>
</ul>
<h3>5. Connect OpenAI</h3>
<ul>
<li>Add your OpenAI API key</li>
<li>Ensure access to chat models and embeddings</li>
<li>No additional configuration required</li>
</ul>
<h3>6. Activate the Workflow</h3>
<ul>
<li>Enable the workflow</li>
<li>Send a test email to your Gmail inbox</li>
<li>Verify that:
<ul>
<li>Email appears in Google Sheets</li>
<li>Categories are assigned</li>
<li>Audio summary arrives in Telegram</li>
</ul>
</li>
</ul>
<hr>
<h2>💡 Recommended Usage</h2>
<ul>
<li>Email triage automation</li>
<li>Customer support inbox classification</li>
<li>Sales lead categorization</li>
<li>Personal productivity systems</li>
<li>AI-assisted knowledge management</li>
</ul>
<hr>
<h2>🧠 Architecture Notes</h2>
<p>
This workflow uses a Retrieval-Augmented Generation (RAG) pattern inside n8n by combining
Google Sheets as a lightweight knowledge base with an in-memory vector store.
Instead of relying purely on model reasoning, classification decisions are grounded
in previously approved examples, producing stable and explainable categorization.
</p>
<p>
Over time, the workflow becomes smarter as new categories are automatically added,
effectively turning your inbox into a continuously learning AI system.
</p>