Add lakebase-devconnect demo script

demos/lakebase-devconnect/Demoscript-operational-lakebase.md

@@ -0,0 +1,84 @@
+# Operational Lakebase Demo Script
+
+## Pre-Demo Checklist
+
+> **CK-yourname workspace** — Confirm everything below is ready. Setup may take up to 30 minutes.
+
+- [ ] **Model Serving Endpoint** — `caspersdev_support_agent` (named "Support Agent")
+- [ ] **Jobs & Pipelines** — `Support Request Agent Stream`
+- [ ] **Catalog Tables**
+  - `support_agent_reports` *(assign a compute resource)*
+  - `support_agent_reports_sync` *(auto-synced with Lakebase)*
+- [ ] **Lakebase Instance** — Autoscaling instance is automatically created
+- [ ] **Data Verification** — `support_agent_report_sync` in Lakebase matches the Lakehouse data
+
+---
+
+## Phase 1: Data Ingestion & AI Processing
+
+1. **Access the Workspace**
+   Start in the Databricks workspace for Casper's Kitchen (the ghost kitchen company).
+
+2. **Review the Catalog**
+   Open Unity Catalog to show where all incoming food order events are streamed.
+
+3. **Demonstrate the Support Agent**
+   - Navigate to the Model Serving endpoint named **"Support Agent."**
+   - Explain that this agent processes incoming support requests to suggest refund amounts and draft customer responses.
+
+4. **Inspect the Processing Notebook**
+   - Go to **Jobs & Pipelines** and locate the **"Support Request Agent Stream"** task.
+   - Show the notebook that uses an OpenAI client to query raw support requests and write processed responses to the `support_agent_reports` table.
+
+5. **Verify Structured Data**
+   Return to Unity Catalog and inspect the `support_agent_reports` table's sample data to show the structured JSON responses generated by the agent.
+
+---
+
+## Phase 2: Operationalizing Data with Lakebase
+
+1. **Explain the Need for Lakebase**
+   The Lakehouse is built for analytics. An OLTP database like Lakebase is required for low-latency operational use cases and application support.
+
+2. **Create a Sync Table**
+   Demonstrate synchronizing data from Unity Catalog to Lakebase by selecting a Postgres database instance and a specific branch. *(One is already created for you.)*
+
+3. **Navigate the Lakebase Dashboard**
+   - Use the app switcher to open the Lakebase dashboard.
+   - Confirm that `support_agent_report_sync` in Lakebase matches the Lakehouse data.
+   - *(You may need to manually assign a compute resource.)*
+
+4. **Demonstrate Instant Branching**
+   - Show the UI's ability to instantly create a **"test branch"** of the production data.
+   - Delete data in the test branch to prove it does **not** affect the production environment.
+
+---
+
+## Phase 3: The Support Console Application (Databricks Apps)
+
+1. **Launch the Support Console**
+   Go to **Compute → Apps** and open the **"Support Console."** Highlight the fast page loading enabled by querying Lakebase instead of the Lakehouse directly.
+
+2. **Process a Support Case**
+   Open a pending case from a user. Review the agent-suggested refund, credits, and draft response.
+
+3. **Interact with the AI**
+   - Demonstrate **"regenerating"** a report by providing the agent with new context (e.g., *"customer is angry on the phone"*).
+   - Show the agent updating its suggestions based on persistent data and new context.
+
+---
+
+## Phase 4: Advanced Agentic Development *(Optional)*
+
+1. **Trigger Code Generation**
+   Open a code editor (e.g., Cursor) with the Casper's Kitchen codebase.
+
+2. **Issue a Natural Language Prompt**
+   Provide a prompt asking for a new feature, such as a **"Flag as Abusive"** button. Explicitly instruct the AI agent to create its own Lakebase development branch and a new "dev app" environment to avoid touching production.
+
+3. **Verify the Dev Environment**
+   - Return to the Lakebase dashboard to show the new branch created by the agent.
+   - Go to **Compute → Apps** to find the newly deployed **"Support Dev"** environment.
+
+4. **Test the New Feature**
+   Open the dev app and demonstrate the **"Handle Abuse"** card to show it works as intended — fully isolated from production.