Forward Deployed Engineer: Building Agentic Fraud Alerting on Databricks

Chapters 0:00 Intro 0:03 What you'll see in this video 1:01 The project at a glance 3:42 Three MCP servers, one risk agent: system map 4:55 The problem: manual fraud review 6:37 Why a forward deployed engineer 8:17 Why MCP: managed, custom and external servers 9:57 What MCP gives us 11:16 Without MCP 12:55 The system prompt 15:21 One transaction, end to end 16:26 From score to action 16:40 Demo: code base and GitHub repo 18:14 Create the workspace folders 19:22 Import the project files 21:06 Run the Unity Catalog setup 23:06 Query the sample data 25:20 Create the Slack channel 26:00 Create the custom MCP server app 27:26 Connect Slack through Unity Gateway 29:11 Deploy the risk-engine app 30:43 Set up a service principal 31:43 Generate the OAuth secret 32:45 Store secrets with the Databricks CLI 33:33 Give the service principal app access 34:30 Verify the MCP connections 35:52 Run the risk agent notebook 36:47 Grant catalog access and load MCP tools 38:48 Warm up and build the agent 39:09 T-9006: HIGH risk, alert sent to Slack 40:36 Low-risk payment: no alert 41:16 T-9004: CRITICAL, payment blocked 43:48 Deep dive: why T-9004 is blocked 45:44 How the risk engine scores it 46:39 The Slack alert and recorded decision 47:15 Scheduled job: preview the pending queue 48:47 VS Code: the job code and databricks.yml 49:36 Deploy and run with the bundle CLI 50:17 Watch the job run 51:36 Job output and Slack alerts 55:28 Run succeeded 55:50 Wrap-up