`), max 100 words. + - Input includes original draft email and human feedback. + - *Input:* Declined feedback and draft email text. + - *Output:* Revised email draft. + - *Edge Cases:* AI rewriting errors, HTML formatting issues. + +--- + +#### 2.3 Knowledge Base Setup and Document Vectorization (Supporting Block) + +**Overview:** +Manages the Qdrant vector database collection and ingests documents from Google Drive, embedding them for retrieval during email response generation. + +**Nodes Involved:** +- When clicking ‘Test workflow’ (Manual Trigger) +- Create collection +- Refresh collection +- Get folder +- Download Files +- Default Data Loader +- Token Splitter +- Embeddings OpenAI1 +- Qdrant Vector Store1 +- Embeddings OpenAI +- Qdrant Vector Store + +**Node Details:** + +- **When clicking ‘Test workflow’** + - *Type:* Manual Trigger + - *Role:* Starts the setup process manually. + - *Input:* None. + - *Output:* Triggers collection creation and refresh. + +- **Create collection** + - *Type:* HTTP Request + - *Role:* Creates a new collection in Qdrant vector database. + - *Configuration:* + - POST to `https://QDRANTURL/collections/COLLECTION` with empty filter. + - Requires HTTP header authentication. + - *Edge Cases:* API errors, authentication failure. + +- **Refresh collection** + - *Type:* HTTP Request + - *Role:* Deletes all points in the Qdrant collection to refresh data. + - *Configuration:* + - POST to `https://QDRANTURL/collections/COLLECTION/points/delete` with empty filter. + - *Edge Cases:* API errors, partial deletion. + +- **Get folder** + - *Type:* Google Drive node + - *Role:* Retrieves files from a specified Google Drive folder (`test-whatsapp`). + - *Configuration:* Uses Google Drive OAuth2 credentials. + - *Edge Cases:* Permission issues, empty folder. + +- **Download Files** + - *Type:* Google Drive node + - *Role:* Downloads each file from the folder, converting Google Docs to plain text. + - *Configuration:* Conversion enabled for Google Docs to `text/plain`. + - *Edge Cases:* Unsupported file types, download failures. + +- **Default Data Loader** + - *Type:* LangChain Document Loader + - *Role:* Loads downloaded files as binary data for embedding. + - *Edge Cases:* Corrupt files, unsupported formats. + +- **Token Splitter** + - *Type:* LangChain Text Splitter (Token-based) + - *Role:* Splits documents into chunks of 300 tokens with 30 token overlap for embedding. + - *Edge Cases:* Incorrect chunk sizes, tokenization errors. + +- **Embeddings OpenAI1** + - *Type:* LangChain Embeddings (OpenAI) + - *Role:* Generates vector embeddings for document chunks. + - *Edge Cases:* API rate limits, embedding failures. + +- **Qdrant Vector Store1** + - *Type:* LangChain Vector Store (Qdrant) + - *Role:* Inserts document embeddings into Qdrant collection. + - *Edge Cases:* API errors, data insertion failures. + +- **Embeddings OpenAI** + - *Type:* LangChain Embeddings (OpenAI) + - *Role:* Generates embeddings for query vectors during retrieval. + - *Edge Cases:* Same as above. + +- **Qdrant Vector Store** + - *Type:* LangChain Vector Store (Qdrant) + - *Role:* Retrieves relevant business knowledge documents for RAG during email reply generation. + - *Edge Cases:* Retrieval failures, empty results. + +--- + +### 3. Summary Table + +| Node Name | Node Type | Functional Role | Input Node(s) | Output Node(s) | Sticky Note | +|-------------------------|--------------------------------------|-------------------------------------------------|-------------------------------|-----------------------------|--------------------------------------------------------------------------------------------------------------| +| Email Trigger (IMAP) | Email Read (IMAP) | Listens for incoming emails | None | Markdown | | +| Markdown | Markdown converter | Converts email HTML to plain text | Email Trigger (IMAP) | Email Summarization Chain | Convert email to Markdown format for better understanding of LLM models | +| Email Summarization Chain| LangChain Summarization Chain | Summarizes email content concisely | Markdown | Write email | Chain that summarizes the received email | +| Write email | LangChain Agent (AI Language Model) | Generates professional reply based on summary | Email Summarization Chain | Edit Fields | Agent that retrieves business information from a vector database and processes the response | +| Edit Fields | Set node | Prepares draft email text for sending | Write email | Gmail | | +| Gmail | Gmail node | Sends draft email for human review and waits | Edit Fields | Text Classifier | IMPORTANT: For the "Send Draft" node, you need to send the draft email to a Gmail address because it is the only one that allows the "Send and wait for response" function. | +| Text Classifier | LangChain Text Classifier | Classifies human feedback as Approved or Declined| Gmail | Send Email, Email Reviewer | Based on the suggestion received, the text classifier can understand whether the feedback received approves the generated email or not. | +| Send Email | Email Send (SMTP) | Sends final approved email to original sender | Text Classifier (Approved) | None | | +| Email Reviewer | LangChain Agent (AI Language Model) | Rewrites email based on human feedback | Text Classifier (Declined) | Edit Fields | The Email Reviewer agent, taking inspiration from human feedback, rewrites the email | +| When clicking ‘Test workflow’ | Manual Trigger | Starts setup process | None | Create collection, Refresh collection | | +| Create collection | HTTP Request | Creates Qdrant collection | When clicking ‘Test workflow’ | Refresh collection | STEP 1: Create Qdrant Collection - Change QDRANTURL and COLLECTION | +| Refresh collection | HTTP Request | Deletes all points in Qdrant collection | Create collection | Get folder | STEP 1: Create Qdrant Collection - Change QDRANTURL and COLLECTION | +| Get folder | Google Drive | Retrieves files from Google Drive folder | Refresh collection | Download Files | STEP 2: Documents vectorization with Qdrant and Google Drive - Change QDRANTURL and COLLECTION | +| Download Files | Google Drive | Downloads and converts files to plain text | Get folder | Qdrant Vector Store1 | STEP 2: Documents vectorization with Qdrant and Google Drive - Change QDRANTURL and COLLECTION | +| Default Data Loader | LangChain Document Loader | Loads downloaded files as binary data | Token Splitter | Qdrant Vector Store1 | | +| Token Splitter | LangChain Text Splitter (Token) | Splits documents into token chunks | Default Data Loader | Default Data Loader | | +| Embeddings OpenAI1 | LangChain Embeddings (OpenAI) | Generates embeddings for document chunks | Token Splitter | Qdrant Vector Store1 | | +| Qdrant Vector Store1 | LangChain Vector Store (Qdrant) | Inserts document embeddings into Qdrant collection| Embeddings OpenAI1 | None | | +| Embeddings OpenAI | LangChain Embeddings (OpenAI) | Generates embeddings for queries | None | Qdrant Vector Store | | +| Qdrant Vector Store | LangChain Vector Store (Qdrant) | Retrieves relevant documents for RAG | Embeddings OpenAI | Write email, Email Reviewer | | +| DeepSeek Chat Model | LangChain Chat Model (DeepSeek) | Alternative AI model for summarization | None | Email Summarization Chain | | + +--- + +### 4. Reproducing the Workflow from Scratch + +1. **Create Email Trigger (IMAP) Node** + - Type: Email Read (IMAP) + - Credentials: Configure IMAP credentials for your email inbox (e.g., `info@n3witalia.com`). + - Parameters: Default options to listen for new emails. + +2. **Add Markdown Node** + - Type: Markdown converter + - Parameters: Set input HTML as `={{ $json.textHtml }}` to convert incoming email HTML to plain text. + +3. **Add Email Summarization Chain Node** + - Type: LangChain Summarization Chain + - Parameters: + - Summarization prompt: "Write a concise summary of the following in max 100 words: \"{{ $json.data }}\" Do not enter the total number of words used." + - Operation mode: Node input binary with key `data`. + - Connect Markdown output to this node. + +4. **Add Write email Node** + - Type: LangChain Agent (AI Language Model) + - Parameters: + - Text: `Write the text to reply to the following email:\n\n{{ $json.response.text }}` + - System message: "You are an expert at answering emails. You need to answer them professionally based on the information you have. This is a business email. Be concise and never exceed 100 words. Only the body of the email, not create the subject." + - Prompt type: Define + - Enable output parser. + - Connect Email Summarization Chain output here. + +5. **Add Edit Fields Node** + - Type: Set node + - Parameters: Assign `email` field to `={{ $json.output }}` (the draft email text). + - Connect Write email output here. + +6. **Add Gmail Node** + - Type: Gmail + - Credentials: Configure Gmail OAuth2 credentials for the review email account (e.g., `info@n3w.it`). + - Parameters: + - Send To: `info@n3w.it` (human reviewer). + - Subject: `=[Approval Required] {{ $('Email Trigger (IMAP)').item.json.subject }}` + - Message: Include original email HTML and AI response: + ``` +
where necessary. Be concise and never exceed 100 words. Only the body of the email." + - Prompt type: Define + - Enable output parser. + - Connect Text Classifier output on "Declined" path here. + +10. **Connect Email Reviewer output back to Edit Fields** + - This allows iterative editing based on human feedback. + +--- + +**Optional: Knowledge Base Setup** + +11. **Add Manual Trigger Node** + - To start setup manually. + +12. **Add Create collection Node** + - HTTP Request POST to `https://QDRANTURL/collections/COLLECTION` with empty filter. + - Authenticate with HTTP header credentials. + +13. **Add Refresh collection Node** + - HTTP Request POST to `https://QDRANTURL/collections/COLLECTION/points/delete` with empty filter. + +14. **Add Get folder Node** + - Google Drive node to list files in folder `test-whatsapp`. + +15. **Add Download Files Node** + - Download and convert Google Docs to plain text. + +16. **Add Default Data Loader Node** + - Load downloaded files as binary data. + +17. **Add Token Splitter Node** + - Split documents into 300 token chunks with 30 token overlap. + +18. **Add Embeddings OpenAI1 Node** + - Generate embeddings for document chunks. + +19. **Add Qdrant Vector Store1 Node** + - Insert embeddings into Qdrant collection. + +20. **Add Embeddings OpenAI Node** + - Generate embeddings for queries. + +21. **Add Qdrant Vector Store Node** + - Retrieve relevant documents for RAG during email reply generation. + +--- + +### 5. General Notes & Resources + +| Note Content | Context or Link | +|-----------------------------------------------------------------------------------------------------------|------------------------------------------------------------------------------------------------| +| For the "Send Draft" node, Gmail is required because it uniquely supports the "Send and wait for response" feature. | Sticky Note6 | +| The workflow supports integration with both Gmail and Outlook via appropriate trigger nodes. | Sticky Note3 | +| Change placeholders `QDRANTURL` and `COLLECTION` when setting up Qdrant collection and vector store nodes.| Sticky Note3, Sticky Note4 | +| The Email Reviewer agent rewrites emails incorporating human feedback using limited HTML formatting. | Sticky Note8 | +| The Text Classifier node helps automate decision-making on whether to send or revise emails based on feedback.| Sticky Note7 | +| For detailed setup, ensure all credentials (IMAP, SMTP, OpenAI API, Gmail OAuth2, Qdrant API, Google Drive OAuth2) are configured properly in n8n.| Workflow Description and Node Credentials sections | + +--- + +This document provides a detailed, structured reference to understand, reproduce, and maintain the "Effortless Email Management with AI-Powered Summarization & Review" workflow, enabling advanced users and AI agents to work effectively with it. \ No newline at end of file