dbt Architect study plan
dbt Architect · Give each week a clear focus.
8 weeks · About 6 hours per week
dbt project architecture and modeling
Diagram the layering strategy (staging → intermediate → marts) for a multi-source warehouse and justify each boundary in writing
Classify ten candidate models as staging, intermediate, or mart and defend the classification against the criteria you wrote
Draft a naming and folder convention document covering sources, staging, marts, and tests
Identify which raw sources need deduplication or type casting before any business logic touches them
Your checkpoint
A one-page architecture decision record containing the layer diagram, naming conventions, and the model classification table
dbt project architecture and modeling
Choose between star-schema dimensional marts and wide denormalized marts for two different reporting use cases and document the tradeoffs
Design incremental materializations for at least two models, specifying the unique key, late-arriving data handling, and lookback window
Write the contract for a mart model: column names, data types, nullability, and accepted values
Trace one business metric from raw source to mart and confirm no fan-out occurs at any join
Your checkpoint
Two fully specified model designs (one dimensional, one incremental) with grain statements, keys, and column contracts
Data platform and environment design
Compare warehouse options (Snowflake, BigQuery, Databricks, Redshift) on concurrency, pricing model, and dbt adapter constraints, then select one for a stated workload
Map the dev, staging, and prod target configuration including schemas, credentials handling, and environment variables
Define how source freshness thresholds and warehouse time zones are set per environment
Specify a schema-per-developer isolation pattern that prevents dev runs from colliding with prod objects
Your checkpoint
An environment design document listing targets, schema naming rules, credential strategy, and freshness thresholds per source
Data platform and environment design & Orchestration, deployment, and CI/CD
Design a branching and promotion strategy from feature branch to production with required checks at each gate
Sequence a CI pipeline that runs on pull requests: lint, compile, run modified models plus their downstream dependents, and test
Configure a slim CI selector so only changed models and their children execute
Define the deployment job that promotes to production and how it differs from the PR job
Your checkpoint
A CI/CD pipeline specification with job stages, selectors, trigger conditions, and a diagram of branch-to-prod flow
Orchestration, deployment, and CI/CD
Compare orchestrator options (Airflow, Dagster, dbt Cloud scheduler, Prefect) for dependency management, retries, and observability, then recommend one
Design a daily run graph showing source freshness checks, model runs, snapshot jobs, and test steps in execution order
Specify retry, alerting, and failure-notification behavior for each job in the graph
Plan a backfill procedure for a three-month historical reload without duplicating rows
Your checkpoint
A run-graph diagram with job order, dependencies, retry policy, and a written backfill runbook
Governance, documentation, and data quality
Write a data quality test strategy covering uniqueness, referential integrity, accepted values, and freshness, and assign each test type to the right layer
Decide which models require contracts and which can stay flexible, with reasoning tied to consumer count
Draft exposure definitions linking downstream dashboards and reports to the models they depend on
Specify a documentation standard for model descriptions, column descriptions, and owner metadata
Your checkpoint
A data quality and governance standard covering test matrix by layer, contract policy, exposure list, and documentation checklist
Performance, cost optimization, and scalability
Identify the top five cost drivers in a warehouse workload and match each to a dbt-level control such as materialization choice, partitioning, or clustering
Redesign two expensive models using partition pruning or incremental filtering and estimate the reduction in scanned bytes
Set up a query-cost monitoring routine that flags models exceeding a byte-scan threshold
Choose the right materialization for a high-churn model and justify it against table and view alternatives
Your checkpoint
A cost optimization report naming the five drivers, the applied changes, and estimated savings per model
Governance, documentation, and data quality & Performance, cost optimization, and scalability
Review the full project against all five domains and close gaps in tests, contracts, and documentation
Run a scalability stress scenario: simulate doubling source volume and note which models break first and why
Assemble a one-page architecture summary tying layers, environments, orchestration, governance, and cost controls together
Take a timed practice set across all domains and log every missed concept with the correct reasoning
Your checkpoint
A consolidated architecture summary plus a scored practice-set review log listing weak areas and remediation notes
Milestones
Week 2: You can state the grain of every model you designed and explain why each join cannot fan out
Week 4: You can describe, without notes, what runs on a pull request versus what runs on a production deploy
Week 6: You can assign a test type to any model layer and justify why contracts apply to some models and not others
Week 8: You can walk through the end-to-end architecture from source to dashboard while naming the cost and quality control at each stage