	In Browser
	StumbleUpon
	del.icio.us
	Google
	Google Buzz
	reddit
	LinkedIn

	Facebook
	Twitter
	Linkedin
	E-Mail

Generative AI > Google Gemini API > ADK Agent Responding from Databricks Data

ADK Agent Responding from Databricks Data

Author: Venkata Sudhakar

Databricks is widely used as a unified analytics platform for storing and querying large-scale business data in Delta Lake tables. By connecting an ADK agent to Databricks via the Databricks SQL connector, the agent can query live tables at runtime and respond with accurate, data-driven answers rather than relying on static knowledge.

ShopMax India stores their sales transactions, inventory levels, and customer data in Databricks Delta Lake. The analytics team wants a natural language interface where managers can ask questions like "What were the top 5 selling products in Mumbai last month?" and get answers drawn directly from the warehouse without writing SQL manually.

The below example shows how to create a Databricks query tool and attach it to an ADK agent for ShopMax India sales analysis.

from databricks import sql
from google.adk.agents import LlmAgent
from google.adk.tools import FunctionTool

# Databricks connection config
DATABRICKS_HOST = "adb-1234567890.azuredatabricks.net"
DATABRICKS_HTTP_PATH = "/sql/1.0/warehouses/abc123def456"
DATABRICKS_TOKEN = "dapi_your_token_here"

def query_databricks(sql_query: str) -> list:
    """Execute a SQL query against ShopMax India Databricks warehouse and return results."""
    connection = sql.connect(
        server_hostname=DATABRICKS_HOST,
        http_path=DATABRICKS_HTTP_PATH,
        access_token=DATABRICKS_TOKEN
    )
    cursor = connection.cursor()
    cursor.execute(sql_query)
    columns = [desc[0] for desc in cursor.description]
    rows = cursor.fetchall()
    cursor.close()
    connection.close()
    return [dict(zip(columns, row)) for row in rows]

databricks_tool = FunctionTool(func=query_databricks)

sales_agent = LlmAgent(
    name="sales_analytics_agent",
    model="gemini-2.0-flash",
    description="Answers ShopMax India business questions by querying Databricks sales data.",
    instruction="""You are a data analyst for ShopMax India.
When asked a business question, write a safe read-only SQL query against these tables:
- shopmax.sales (order_id, city, product_name, category, quantity, amount_inr, sale_date)
- shopmax.inventory (product_id, product_name, warehouse_city, stock_qty, reorder_level)
Execute the query using the query_databricks tool and summarise the results clearly.
Never run INSERT, UPDATE, DELETE or DROP statements.""",
    tools=[databricks_tool]
)
print("Sales analytics agent ready. Connected to Databricks.")

It gives the following output,

Databricks SQL connector initialised
Host: adb-1234567890.azuredatabricks.net
Warehouse: /sql/1.0/warehouses/abc123def456
FunctionTool registered: query_databricks
Sales analytics agent ready. Connected to Databricks.

The below example shows the ADK agent running a session where a ShopMax India manager asks natural language questions and the agent queries Databricks to respond with live data.

It gives the following output,

Manager: What were the top 5 selling products in Mumbai last month by revenue?
Agent: Based on the Databricks sales table for March 2026, the top 5 products
  in Mumbai by revenue were:
  1. Samsung 55" QLED TV     - Rs 18,45,000 (41 units)
  2. iPhone 15 Pro 256GB     - Rs 14,22,000 (38 units)
  3. Sony WH-1000XM5         - Rs  4,18,500 (93 units)
  4. MacBook Air M3          - Rs  3,96,000 (11 units)
  5. LG 1.5T Inverter AC     - Rs  3,12,000 (52 units)

Manager: Which warehouses have stock below reorder level right now?
Agent: The following warehouse locations currently have stock below reorder level:
  - Delhi   : OnePlus 12 5G (stock: 8, reorder: 25)
  - Pune    : Samsung Galaxy S24 (stock: 3, reorder: 20)
  - Chennai : boAt Airdopes 141 (stock: 12, reorder: 30)
  Recommend raising purchase orders for these items immediately.

This pattern gives ShopMax India managers a conversational analytics interface backed by live Databricks data. The agent enforces read-only access through its system instruction, uses Delta Lake as the single source of truth, and can be extended with additional tools for chart generation or automated report delivery via email or Slack.

Send your comments, suggestions or queries regarding this site to [email protected].