Skip to content

Level 4 · Building, Scaling & Operating

Level 4 is about building on Snowflake and running it well: code that runs next to the data, AI functions in SQL, and the operational disciplines (cost, scaling, continuity, deployment, migration) that decide whether a platform stays healthy as it grows. Along the way you'll see a "vectorized" handler run more than twice as slowly as the plain one, label descriptions multiply an AI function's input tokens about six times, 86% of a warehouse bill attributed to nobody, four small clusters beat a larger warehouse when queries don't scale with size, a dbt incremental model pass every test while serving stale revenue, and translated SQL that compiles but returns different answers.

What was run

Nothing here was run on Snowflake. Snowpark code ran in Snowpark's own local testing mode (Snowpark 1.53.0), a dbt project ran with dbt-core 1.12.5 on DuckDB 1.5.6, UDF handlers and a UDTF ran as plain Python, and the capstone's chargeback query ran in DuckDB with pytest and a mutation check. Costs, scaling and data loss come from small models with their assumptions stated. All Snowflake SQL was syntax-checked and verified against the October 2026 documentation, including features that are new in 2025–2026: Cortex AI functions and model access control, Adaptive Warehouses, Gen2 as the default, Optimized Refresh and RPO Assurance, and dbt Projects on Snowflake.

Prerequisites: Levels 1–3, and Python for lessons 02, 03 and 10.

Modules

  1. UDFs & Stored Procedures — SQL UDFs and table functions, Snowflake Scripting with transactions, and owner's vs caller's rights.
  2. Snowpark for Python — lazy DataFrames pushed down as SQL, the to_pandas trap, pandas on Snowflake, and unit tests in local testing mode.
  3. Python UDFs, Vectorized UDFs & UDTFs — runtimes and packages, external access, when batching helps (measured), and a sessionizing table function.
  4. Cortex AI Functions — classify, filter, sentiment and structured extraction in SQL, access control, data residency and token costs.
  5. Cost Management: Monitors, Budgets & Attribution — where credits go, the idle gap in per-query attribution, tags, resource monitors and budgets.
  6. Warehouse Scaling: Up, Out & Adaptive — scale-up vs scale-out modelled, multi-cluster policies, query acceleration, Gen2 and Adaptive Warehouses.
  7. Replication, Failover & Business Continuity — replication and failover groups, refresh modes, data loss modelled, Client Redirect and a runbook.
  8. dbt & CI/CD for Snowflake — a tested dbt project, an incremental bug only a rebuild comparison caught, clones per pull request, and dbt Projects on Snowflake.
  9. Migrating to Snowflake — translation and its semantic traps, moving data, and reconciliation that catches export bugs.
  10. Capstone — A Cost-Aware Data Platform — owned warehouses, monitors and budgets, and a tested monthly chargeback that assigns every credit.

After Level 4

You've covered Snowflake from architecture to operations. Good next steps: run the capstone in a trial account with your own workloads, take the dbt and data engineering path further, or pair this course with the PySpark Mastery Path to compare the two platforms on the same pipeline.