It’s all about data.

As AI continues to evolve beyond traditional chatbots and Q&A systems, the real frontier lies in agent systems—intelligent assistants that not only answer questions but take actions, invoke tools, and automate tasks. With the recent partnership between Anthropic and Databricks, it’s now possible to integrate the powerful Claude 3.7 Sonnet model directly within your Databricks workspace. Even better—Databricks has made it easier than ever to build these agents natively, using your own data and custom tools. In this guide, we’ll build a smart Marketing Assistant Agent that: Retrieves customer data Uses vector search to find product descriptions Generates a personalized marketing email Sends it out using AWS SES All using Databricks, Claude 3.7, and a few simple SQL and Python functions. Step 1: Set Up Your Free Databricks Account To get started, head over to signup.databricks.com and choose the Express setup. You’ll receive some free credits and get access to a fully functional serverless Databricks workspace—no need to set up a cloud account, networking, or storage configurations. 💡 Tip: The Playground is your testbed for experimenting with models like Llama or Claude, and for registering functions (also known as tools) that your agent can use during execution. Once you're in, navigate to Serving → Playground and you should see the native integration with Claude 3.7 available in the model selector. ⚠️ Don’t see Claude 3.7? That’s likely because your workspace is not hosted in a US region—Claude is currently only available in the US. As a temporary solution for testing—or if your company policy permits—you can disable cross-geo processing in your workspace settings. For detailed instructions, refer to this documentation. Native Claude hosting in European regions is expected soon, which will remove this limitation. Step 2: Define the Agent Tools in Unity Catalog The core idea of agent systems in Databricks is simple: you create Python or SQL functions, register them in Unity Catalog, and make them available to your agent via the Playground. In our case, we’ll implement three tools that the Claude agent will automatically call to complete its workflow: End-to-End Solution Architecture Let’s define these three functions in Unity Catalog: CREATE OR REPLACE FUNCTION agent.claude.lookup_customer_info( input_customer_name STRING COMMENT 'Name of the customer whose info to look up' ) RETURNS STRING COMMENT "Returns metadata about a particular customer given the customer name, including the customer's email, ID and last ordered item. The customer ID can be used for other queries." RETURN SELECT CONCAT( 'Customer ID: ', customer_id, ', ', 'Customer Email: ', customer_email, ', ', 'Last Ordered Item: ', last_ordered_item ) FROM agent.claude.customer_info WHERE customer_name = input_customer_name LIMIT 1; ✅ This function allows Claude to dynamically retrieve a customer's email and last ordered item based on their name. 2.2. product_description_similarity_search: Enrich product descriptions using vector search CREATE OR REPLACE FUNCTION agent.claude.product_description_similarity_search ( query STRING COMMENT 'The query string for searching product documents.' ) RETURNS TABLE COMMENT 'Executes a search on product documents to retrieve text documents most relevant to the input query.' RETURN SELECT chunk_text as page_content, id as metadata FROM vector_search( index => 'agent.claude.product_description_vs_index', query => query, num_results => 5 ); 💡 This uses the native Databricks vector search to find relevant product details, helping Claude personalize the email. Creating a vector search index in Databricks 2.3. send_email_via_ses: Automatically send the email CREATE OR REPLACE FUNCTION agent.claude.send_email_via_ses( recipient_email STRING COMMENT 'Recipient email address', subject STRING COMMENT 'Subject of the email.', message STRING COMMENT 'Plain text content for the email body.' ) RETURNS STRING LANGUAGE PYTHON AS $$ import boto3 def send_email_via_ses(recipient_email, subject, message): aws_access_key = "ABC" aws_secret_key = "DEF" aws_region = "eu-central-1" ses_client = boto3.client( "ses", aws_access_key_id=aws_access_key, aws_secret_access_key=aws_secret_key, region_name=aws_region ) sender_email = "thomas.hass@datamyself.ai" html_template = f""" {subject} {message} Shop Now Thanks for being a valued part of our community. """ response = ses_client.send_email( Source=sender_email, Destination={"ToAddresses": [recipient_email]}, Message={ "Subject": {"Data": subject}, "Body": { "Text": {"Data": "Please view this in HTML format."}, "Html": {"Data": html_template} } } ) return response["MessageId"] return send_email_via_ses(recipient_email, subject, message) $$; 🚨 In production, move credentials to Databricks Secrets or AWS Secrets Manager. Step 3: Configure the Agent in Playground Now that our tools are ready, test everything in the Databricks Agent Playground. Select model: Claude 3.7 Sonnet Add tools: agent.claude.* Add a system prompt describing the workflow You are a marketing assistant specializing in personalized campaigns. Step 1: Retrieve customer metadata Step 2: Perform product vector search Step 3: Generate and send an HTML email using SES Example Prompt: Send a marketing email to Thomas highlighting the benefits of his last viewed product and include a 10% discount using code WELCOME10. Claude agent executing multi-step tasks end-to-end 📺 Full YouTube Walkthrough https://www.youtube.com/watch?v=U-eRydmb07g Let me know in the comments if you've built something similar or how you're planning to use Claude-powered agents in your workflows!

It’s all about data.

Building a Marketing Assistant Agent on Databricks using Claude 3.7

How to Merge your Relational Database into a Data Lake on AWS