Overview

This archive is a self-contained quickstart for LangGraph Agent on Databricks. The agent has access to Tavily for web search to assist in Q&A.

Files

• agent_config.yaml: Contains key parameters for the LangGraph, including:
  • Model params: model endpoint names, temperature, max_tokens, etc.
  • Databricks resource params: catalog, schema
• 01_langgraph_agent Contains code for the LangGraph agent and Tavily tool. This notebook should only include model code and nothing else (no testing code, debug statements, etc,)
• 02_evaluate_&_deploy: Tests/runs the code in 01_langgraph_agent. It contains code that:
  • Logs the chain to MLflow
  • Implements mlflow.evaluate() to evaluate agent against a benchmark dataset
  • Registers the chain to Unity Catalog
  • Deploys the chain to a serving endpoint and starts a review UI

Development Workflow

1. Update agent_config.yaml with the Databricks Resources (catalog, schema), model resources (model endpoints, temperature, max tokens, etc.)
2. Review and customize 01_langgraph_agent as needed.
   • Note: This notebook should not be run indepedently. To test your code changes and customizations, load this notebook in 02_evaluate using %run, and then invoke the model in the 02_evaluate notebook
3. Test the agent code in 02_evaluate. Once the code is stabilized, log the model, run evaluations, register, and deploy.

Requirements

• Permissions to write to schemas in Databricks
• Permissions to deploy model serving endpoints
• Enablement of AI-assisted features on your Databricks workspace