MLflow Prompt Registry, Persistent Memory, Databricks Apps | Databricks GenAI Engineer Cert

Master the production deployment stack for GenAI agents: MLflow Prompt Registry for governed prompt lifecycle and gated release, Lakebase for persistent agent memory, and Databricks Apps as the OAuth-native UI layer. This video covers three critical capabilities that close deployment gaps and appear directly on the exam. You'll learn how to version and promote prompts without redeploying code, implement stateful agents with durable memory across restarts, and build secure agent UIs with proper OAuth authentication—not PATs in JavaScript. The video walks through the DataDino case study: a regression in exhibit descriptions that required rollback, solved by Prompt Registry aliases. • MLflow Prompt Registry: versions, aliases, and the prompts:// URI grammar • Gated release with immutable versions and mutable aliases (dev → staging → prod) • Lakebase: managed Postgres for short-term agent memory and Delta for long-term chat history • Databricks Apps: service principals, on-behalf-of OAuth, and why PATs in JavaScript leak tokens • Declarative Automation Bundles for promoting indexes, prompts, and agents across environments • Sample Q8: internal RAG app with corporate SSO and the hard rule on OAuth vs PAT 0:00 Introduction 0:35 Previously On 1:51 The Three Deployment Gaps 3:12 MLflow Prompt Registry & Gated Release 5:19 Versions vs Aliases 6:26 Why Apps Load Aliases 8:34 Git-Versioned Prompts vs Registry 9:34 Problem Two: Agent Memory 10:11 Lakebase and Persistent State 11:15 Short-Term vs Long-Term Memory 11:54 The Stateful Agent Pattern 14:09 Problem Three: Roll-Your-Own Flask UIs 14:46 Databricks Apps as the Solution 15:35 Service Principal vs On-Behalf-Of OAuth 18:20 Sample Q8: Internal RAG App with Corporate SSO 19:40 The Hard Rule: No PATs in JavaScript 20:04 Why a PAT in JavaScript Is a Real Security Failure 22:04 Databricks Asset Bundles for CI/CD 22:25 Why Bundles Exist 23:26 Promoting With a Bundle 25:13 Decision Rules 25:57 Three Legs of the Production Stool 🔗 Databricks GenAI Engineer Exam: https://www.databricks.com/learn/cert... 📋 Full Playlist: https://www.youtube.com/@StackLessons... StackLessons creates hands-on exam prep content for cloud & AI certifications. Like & Subscribe for more content! Need more practice questions? Visit https://certcompanion.com/exams #Databricks #GenAI #MLflow 📌 Chapters 0:00 Introduction 0:35 Previously On 1:51 The Three Deployment Gaps 3:12 MLflow Prompt Registry & Gated Release 5:19 Versions vs Aliases 6:26 Why Apps Load Aliases 8:34 Git-Versioned Prompts vs Registry 9:34 Problem Two: Agent Memory 10:11 Lakebase and Persistent State 11:15 Short-Term vs Long-Term Memory 11:54 The Stateful Agent Pattern 14:09 Problem Three: Roll-Your-Own UIs 14:46 Databricks Apps as the Solution 15:35 Service Principal vs On-Behalf-Of OAuth 18:20 Sample Q8: Internal RAG App with SSO 19:40 The Hard Rule: No PATs in JavaScript 20:04 Why PATs in JavaScript Leak Tokens 22:04 Databricks Asset Bundles for CI/CD 23:26 Promoting With a Bundle 25:13 Decision Rules 25:57 Three Legs of the Production Stool #Databricks #GenAI #MLflow