Databricks Certified Generative AI Engineer Associate Course

This course contains the use of artificial intelligence.

Prepare for the Databricks Certified Generative AI Engineer Associate exam with a complete video course built around the official exam guide. The exam tests whether you can design, build, deploy, govern, and evaluate generative AI applications on the Databricks Data Intelligence Platform, and this course covers every one of its six sections, with time spent in proportion to the exam's weights.

The course runs about 4.3 hours across 144 lectures in 13 sections. Instead of a tour of features, it follows one fictional company, Harborline Insurance, as its team builds PolicyPal, a policy assistant that answers customer and agent questions from thousands of policy documents, looks up claims, respects each user's permissions, and is evaluated and monitored in production. Every decision you'll face on the exam appears in that story, with the reasoning behind it.

What makes this course different

Mapped to the exam guide. Every section names the exam section it covers, so you always know where you are and how much that section is worth.

Real code, explained line by line. Python and SQL examples use ai_parse_document, Delta tables, Vector Search, LangChain, MLflow ResponsesAgent, Agent Framework deployment, ai_query, and MLflow GenAI evaluation.

Decisions, not definitions. Lectures focus on when to choose one option over another: a chain or an agent, Vector Search or Genie or a Unity Catalog function, pay-per-token or provisioned throughput, standard or storage-optimized endpoints, a code scorer or an LLM judge.

Exam-style practice everywhere. Each section ends with scenario questions where every option is explained. A dedicated exam traps section compares look-alike features side by side, and the practice sets include more than 100 exam-style questions and four longer case scenarios.

A complete build lab. One section follows PolicyPal from raw PDF files in a volume to a monitored endpoint: parsing, chunking, incremental updates, indexing, the agent and its tools, logging and registration, deployment, evaluation, and monitoring.

What's covered

Section 1: Design Applications (14%). Matching model tasks to requirements, prompt anatomy, prompts that return a specific format, structured outputs, chain components, tools for multi-stage reasoning, agent planning, and Agent Bricks.

Section 2: Data Preparation (14%). Choosing source documents, extracting content by format, ai_parse_document, filtering noise, chunking by size, structure, and model limits, advanced chunking, writing chunks to Delta tables, retrieval evaluation, and re-ranking.

Section 3: Application Development (30%). LangChain and LangGraph, query rewriting and intent routing, prompt techniques, quality and safety issues, guardrails, selecting LLMs and embedding models, model cards and Marketplace, Agent Framework and MLflow Tracing, and multi-agent systems with Genie spaces.

Section 4: Assembling and Deploying Applications (22%). Coding chains, pyfunc models, model signatures, registering to Unity Catalog, serving agents and the review app, Foundation Model APIs, access to resources from endpoints, ai_query and batch inference, Vector Search concepts, indexes, queries, and configuration, agent memory, MCP, Prompt Registry, CI/CD, and user interfaces with Databricks Apps.

Section 5: Governance (8%). Masking, defending against malicious inputs, legal and licensing requirements, and mitigating problematic text in data sources.

Section 6: Evaluation and Monitoring (12%). Choosing models from metrics, MLflow GenAI evaluation, built-in judges and ground truth, custom scorers, inference tables, production monitoring, key metrics, AI Gateway usage tracking and rate limits, cost control, and subject matter expert feedback.

Course structure

1. Welcome and orientation. The exam format, sections and weights, question styles, the Harborline case study, and a study plan.

2. Design applications. Tasks, prompts, structured output, chains, tools, agents, and Agent Bricks.

3. Data preparation. Sources, parsing, filtering, chunking, Delta tables, retrieval evaluation, and re-ranking.

4. Prompts, quality, and guardrails. Frameworks, prompt augmentation, routing, prompt techniques, quality and safety issues, and guardrails.

5. Models, agents, and Genie. Model selection, embeddings, Agent Framework, tracing, evaluation and monitoring in the lifecycle, and Genie.

6. Chains, registration, and serving. Chains, pyfunc, signatures, Unity Catalog registration, deployment, Foundation Model APIs, access control, and ai_query.

7. Vector Search and production. Indexes, queries, configuration, memory, MCP, Prompt Registry, CI/CD, and user interfaces.

8. Governance. Masking, malicious inputs, licensing, and problematic source text.

9. Evaluation and monitoring. Judges, scorers, inference tables, production monitoring, AI Gateway, cost, and expert feedback.

10. Build lab. PolicyPal end to end, from documents to a monitored endpoint.

11. Exam traps. Side-by-side comparisons of features that candidates often confuse.

12. Practice sets. Twenty sets of exam-style questions across all six exam sections, each answer explained.

13. Final review. Four case scenarios, a rapid review of every exam section, and exam-day strategy.

Practice and assessment

Section quizzes check your understanding as you go. The course ends with a full-length practice exam of 45 original questions in the real exam's style, including multiple-select questions, with an explanation for every option and a reference to the relevant documentation. The practice questions are written fresh for this course and don't repeat the lecture scenarios.

How to use this course

Watch one section at a time, try the code in your own Databricks workspace where you can, take the section quiz, and revisit anything you miss. Use the exam traps and practice sets in the final week, then take the full practice exam under timed conditions. The rapid review lectures give you a compact recap of every exam section for the day before your exam.

Databricks renames and updates features often. Mosaic AI Vector Search is now called AI Search, and AI Gateway is now part of Unity Gateway. The course teaches the names used in the exam guide and points out the newer names, so you'll recognize both.

Disclaimer

This course is not affiliated with, endorsed by, or sponsored by Databricks, Inc. Databricks, Delta Lake, MLflow, Unity Catalog, and related names are trademarks of their respective owners. The exam content and guide belong to Databricks and may change; always check the official exam page before you sit the exam. Harborline Insurance and all people and companies in the examples are fictional.