AI Lead Scoring from MLS Data via Webhook

Summary

This workflow automates lead scoring using MLS (Multiple Listing Service) data, combining vector search and a conversational AI agent. It receives lead or property data via a webhook, processes it into embeddings, stores it in a Pinecone vector database, and then uses a Hugging Face chat model to analyze or score the leads. The results are appended to a Google Sheet for easy tracking and further action.

Ideal for real estate professionals, this workflow turns raw MLS data into actionable intelligence, enabling quick identification of high-value leads, property matching, or automated responses.

Node-by-Node Breakdown

• Webhook (No auth) — Listens for incoming HTTP POST requests at path /lead_scoring_with_mls_data. This is the trigger that starts the workflow whenever new data is sent.
• Splitter (No auth) — A Character Text Splitter that breaks incoming text into chunks of 400 characters with 40-character overlap. This ensures long MLS descriptions are properly segmented for embedding.
• Embeddings (API Key auth) — Generates vector embeddings using OpenAI's default model. Requires an OpenAI API key configured in n8n credentials.
• Insert (API Key auth) — Inserts the generated embeddings into a Pinecone index named lead_scoring_with_mls_data. This creates a searchable vector store of MLS data.
• Query (API Key auth) — Queries the same Pinecone index to retrieve relevant documents. This node shares credentials with the Insert node and is used by the tool.
• Tool (API Key auth) — A Vector Store Tool that wraps the Pinecone query node, making it available to the AI agent as a callable function.
• Memory (No auth) — Buffer Window Memory that stores recent conversation context, allowing the agent to maintain state across interactions.
• Chat (API Key auth) — Hugging Face Chat Model (e.g., Llama, Mistral) used as the language model for the agent. Requires a Hugging Face API key.
• Agent (No auth) — An AI Agent that orchestrates the chat model, the vector store tool, and memory. It processes incoming data (from the webhook), decides when to query the vector store, and generates a response (e.g., a lead score).
• Sheet (OAuth2) — Google Sheets node that appends the agent's output to a sheet named "Log" in the document identified by SHEET_ID. You must replace SHEET_ID with your actual Google Sheet ID and configure OAuth2 credentials for Google Sheets.

Setup Instructions

1. Webhook: No setup required; the webhook URL will be generated when you activate the workflow in n8n.
2. OpenAI: Create an account at platform.openai.com and generate an API key. Add this as a credential in n8n (OpenAI type).
3. Pinecone: Sign up at pinecone.io, create an index named lead_scoring_with_mls_data (dimensions matching OpenAI embeddings, e.g., 1536). Obtain your API key and environment. Add as a Pinecone credential in n8n.
4. Hugging Face: Create an account at huggingface.co and generate an access token. Add as a credential in n8n (Hugging Face type).
5. Google Sheets: Enable the Google Sheets API in your Google Cloud project, create OAuth2 credentials (Desktop app type), and add them in n8n. Replace the placeholder SHEET_ID in the Sheet node with the actual ID of your Google Sheet.
6. Activate the workflow and send a POST request to the webhook URL with your MLS data (e.g., JSON with property details).

Use Cases & Variations

• Real Estate Lead Scoring: Incoming MLS leads are automatically analyzed and scored based on criteria like price, location, or features. The agent can output a score (0–100) and store it in Google Sheets.
• Property Matching: Instead of scoring, the agent can find similar properties from the vector store and recommend them.
• Multi-step Analysis: Add more tools (e.g., a calculator or CRM lookup) to enrich the agent's capabilities.
• Different Trigger: Replace the webhook with a schedule or event-driven trigger (e.g., new email or form submission).
• Alternative Vector DB: Swap Pinecone for Qdrant, Weaviate, or Supabase by changing the vector store type.
• Model Choice: Use OpenAI's chat model instead of Hugging Face by swapping the Chat node.

{
  "name": "Mortgage Rate Alert",
  "nodes": [
    {
      "parameters": {
        "content": "## Mortgage Rate Alert",
        "height": 520,
        "width": 1100
      },
      "id": "e7f69fa7-a800-43bc-b113-061d37523f4d",
      "name": "Sticky",
      "type": "n8n-nodes-base.stickyNote",
      "typeVersion": 1,
      "position": [
        -500,
        -250
      ]
    },
    {
      "parameters": {
        "httpMethod": "POST",
        "path": "mortgage_rate_alert"
      },
      "id": "3b8e8fe5-3986-44f2-9306-790e2ae3119f",
      "name": "Webhook",
      "type": "n8n-nodes-base.webhook",
      "typeVersion": 1,
      "position": [
        -300,
        0
      ]
    },
    {
      "parameters": {
        "chunkSize": 400,
        "chunkOverlap": 40
      },
      "id": "dc561d99-0499-4c81-a7e5-e59ced66a2d5",
      "name": "Splitter",
      "type": "@n8n/n8n-nodes-langchain.textSplitterCharacterTextSplitter",
      "typeVersion": 1,
      "position": [
        -100,
        0
      ]
    },
    {
      "parameters": {
        "model": "default"
      },
// ... truncated (copy to see full JSON)