← Agentic workflow consulting
Certification preparation

Databricks certification consulting

We provide preparation consulting for Databricks certification — structured study plans, hands-on walkthroughs on a real workspace, and targeted drilling on the domains that actually carry weight in the exam. The work is built around the official exam guides rather than question dumps, so what you learn survives the test.

Two tracks are covered in depth. Each breakdown below is reproduced from the current published Databricks exam guide, with the section weightings shown wherever Databricks publishes them.

Part 1

Data Engineer Associate

7 exam sections

Part 2

Generative AI Engineer Associate

6 exam sections

Part 1

Databricks Certified Data Engineer Associate

Foundational data engineering on the Databricks Data Intelligence Platform: ingestion, transformation, orchestration with Lakeflow Jobs, CI/CD, and Unity Catalog governance.

Scored questions
45 multiple choice
Time limit
90 minutes
Registration fee
USD 200
Prerequisite
None; ~6 months hands-on recommended
Validity
2 years

On weightings: Databricks does not publish per-section weightings for the exam version effective 4 May 2026. Objective counts are shown instead as a rough proxy for emphasis — treat them as indicative, not official. Older weightings circulating online (24% / 29% / 22% / 16% / 9%) belong to the retired five-section version and no longer apply.

01Databricks Intelligence Platform

2 objectives
  • ·Core components of the Data Intelligence Platform — architecture, Delta Lake, and Unity Catalog
  • ·Compute services: characteristics, limitations, cost models, and selecting the right option per workload

02Data Ingestion and Loading

7 objectives
  • ·Ingestion patterns — batch, streaming, and incremental loading from local files and Lakeflow Connect connectors
  • ·COPY INTO for incremental loads from ADLS/S3/GCS into Unity Catalog–governed tables
  • ·Auto Loader with schema enforcement and schema evolution in batch modes
  • ·Configuring Lakeflow Connect to ingest from diverse enterprise sources
  • ·JDBC/ODBC and REST clients in notebooks, orchestrated with Lakeflow Jobs
  • ·Choosing between Auto Loader, Lakeflow Connect, and partner connectors by volume, frequency, data type, and governance need
  • ·Ingesting semi-structured and unstructured data (JSON, nested) into Delta tables

03Data Transformation and Modeling

7 objectives
  • ·Bronze-to-silver cleaning with PySpark/SQL — nulls, type standardization, writes
  • ·Joins and set operations: inner, left, broadcast, multi-key, cross, union / union all
  • ·Column, row, and structure manipulation — add, drop, split, rename, filter, explode
  • ·Deduplication and aggregations: count, approx count distinct, mean, summary
  • ·Tuning parameters — shuffle partitions, default parallelism, executor/driver memory, broadcast join threshold
  • ·Gold-layer objects: materialized views, views, streaming tables, and tables for BI
  • ·Data quality checks and validation rules for reliable silver and gold datasets

04Working with Lakeflow Jobs

4 objectives
  • ·Control flow — retries and conditional tasks including branching and looping
  • ·Configuring notebook, SQL query, dashboard, and pipeline tasks with DAG-based dependencies
  • ·Job schedules and trigger types: scheduled, file arrival, and table update
  • ·Choosing time-based versus data-driven triggers based on data availability

05Implementing CI/CD

4 objectives
  • ·Code workflow in the workspace UI — Databricks Repos branches, commits, and merges
  • ·Environment-specific configuration via Automation Bundle variables and overrides
  • ·Deploying Declarative Automation Bundles (formerly Databricks Asset Bundles) across dev, test, and prod
  • ·Databricks CLI for validating, deploying, and managing bundles in automated workflows

06Troubleshooting, Monitoring, and Optimization

5 objectives
  • ·Job performance trends via Lakeflow Jobs run history against historical baselines
  • ·Pipeline health monitoring — job statuses, DAG task graphs, run times, failure rates
  • ·Diagnosing bottlenecks — data skew, shuffling, disk spilling — from Spark UI stage metrics
  • ·Liquid Clustering and predictive optimization
  • ·Cluster startup failures, library conflicts, and out-of-memory issues

07Governance and Security

4 objectives
  • ·Managed versus external tables in Unity Catalog, including converting between them
  • ·Access control via UI and SQL — GRANT, REVOKE, DENY across the security hierarchy
  • ·Column-level masking and row-level security by user group
  • ·Unity Catalog ABAC policies for centralized row filtering and column masking

Effective 4 May 2026 · Official Databricks exam guide →

Part 2

Databricks Certified Generative AI Engineer Associate

Building and shipping LLM applications on Databricks: RAG design, chunking and retrieval, Mosaic AI Vector Search, MLflow and Agent Framework, model serving, guardrails, and evaluation.

Scored questions
45 multiple choice
Time limit
90 minutes
Registration fee
USD 200
Prerequisite
None; ~6 months hands-on recommended
Languages
Python for ML code; SQL for data tasks

01Design Applications

14%
  • ·Designing prompts that elicit a specifically formatted response
  • ·Selecting model tasks to meet a given business requirement
  • ·Selecting chain components for a desired model input and output
  • ·Translating business goals into desired inputs and outputs for an AI pipeline
  • ·Defining and ordering tools for multi-stage reasoning
  • ·When to use Agent Bricks — Knowledge Assistant, Multiagent Supervisor, Information Extraction

02Data Preparation

14%
  • ·Chunking strategy for a given document structure and model constraints
  • ·Filtering extraneous content that degrades RAG quality
  • ·Choosing the right Python package to extract document content by source format
  • ·Writing chunked text into Delta Lake tables in Unity Catalog
  • ·Identifying source documents with the necessary knowledge and quality
  • ·Tools and metrics for evaluating retrieval performance
  • ·Advanced chunking strategies for retrieval system design
  • ·The role of re-ranking in information retrieval

03Application Development

30%
  • ·Selecting LangChain or similar tooling for a GenAI application
  • ·Qualitatively assessing responses for quality and safety issues
  • ·Chunking strategy driven by model and retrieval evaluation
  • ·Augmenting prompts with context from user input — key fields, terms, intents
  • ·Adjusting an LLM's response from a baseline to a desired output
  • ·Implementing LLM guardrails to prevent negative outcomes
  • ·Selecting the best LLM for the application's attributes
  • ·Embedding model context length by source documents, expected queries, and optimization strategy
  • ·Selecting models from a hub or marketplace using metadata and model cards
  • ·Selecting a model from common experiment metrics
  • ·MLflow and Agent Framework for developing agentic systems
  • ·Comparing evaluation and monitoring phases of the GenAI lifecycle
  • ·Multi-agent systems using Genie Spaces or the conversational API

04Assembling and Deploying Applications

22%
  • ·Coding a chain as a pyfunc model with pre- and post-processing
  • ·Controlling access to resources from model serving endpoints
  • ·Core RAG elements — model flavor, embedding model, retriever, dependencies, signature
  • ·Registering models to Unity Catalog with MLflow
  • ·Creating and querying a Vector Search index
  • ·Serving LLM applications with Foundation Model APIs
  • ·Key concepts and components of Mosaic AI Vector Search
  • ·Batch inference workloads and appropriate use of ai_query()
  • ·Configuring vector search by embedding count, update frequency, latency, and cost
  • ·Persistent datastores for intermediate memory or structured information
  • ·CI/CD practice — updating indexes, promoting prompts across environments, component testing
  • ·Integrating managed, external, and custom MCP servers
  • ·Prompt version control and lifecycle management
  • ·User-facing interfaces for agents — Apps, Slack, Teams

05Governance

8%
  • ·Masking techniques as guardrails against a performance objective
  • ·Guardrail techniques protecting against malicious user input
  • ·Legal and licensing requirements for data sources
  • ·Alternatives for problematic text in a source feeding a GenAI application

06Evaluation and Monitoring

12%
  • ·Selecting LLM size and architecture from quantitative evaluation metrics
  • ·Key metrics to monitor for a given deployment scenario
  • ·Evaluating agent performance with MLflow scoring and tracing
  • ·Inference logging to assess deployed RAG performance
  • ·Databricks features for controlling LLM costs
  • ·Inference tables and Agent Monitoring for live endpoints
  • ·Identifying evaluation judges that require ground truth
  • ·AI Gateway — inference tables, usage tables, rate limiting
  • ·Custom Scorers for evaluating agents and LLMs
  • ·Incorporating SME feedback to improve agent performance

March 2026 guide · Official Databricks exam guide →

Preparing a team, or yourself?

Tell us where you are — first attempt, retake, or rolling a whole team through certification — and we’ll map a plan against the sections above.

Start a conversation

Databricks and the certification names above are trademarks of Databricks, Inc. This is independent preparation consulting and is not affiliated with or endorsed by Databricks.