InternshipAll 30 tasks, level by level
The real graded Databricks & Snowflake tasks the portal assigns, grouped into three levels.
The portal assigns these Databricks & Snowflake tasks. Finish Level 1 before Level 2 and Level 2 before Level 3. Your mentor may mark some tasks as required for the certificate.
Level 1: Foundations (12 tasks)
| Task | What to do |
|---|---|
| Warehouses and lakehouses | In a repo named lakehouse-internship, explain data warehouses, data lakes and lakehouses, OLTP versus OLAP, and ETL versus ELT with examples. |
| Free accounts | Sign up for Databricks Free Edition and a Snowflake 30-day trial. Set Snowflake warehouse auto-suspend to 60 seconds and a resource monitor. Screenshot the setup. |
| SQL refresher | Write 10 SQL queries on a sample sales dataset: filters, joins, GROUP BY, HAVING, CASE and subqueries. |
| Window functions | Write queries using ROW_NUMBER, RANK, LAG and running totals to find each customer's latest order and month-on-month growth. |
| Load a CSV into Snowflake | Create a database, schema and table, upload a CSV to a stage and load it with COPY INTO. Handle a bad row on purpose. |
| Snowflake warehouses | Run the same query on X-Small and Small warehouses, compare time and credits, and explain when to size up. |
| Databricks notebook basics | In a Databricks notebook, read a CSV with PySpark, show the schema and run 5 DataFrame operations (select, filter, groupBy, join, orderBy). |
| Spark SQL | Register a DataFrame as a view and answer the same 5 questions with Spark SQL. |
| Parquet versus CSV | Save the same data as CSV and Parquet, compare file size and query time, and explain columnar storage. |
| Create Delta tables | Write your data as a Delta table, run an UPDATE and a DELETE, and use DESCRIBE HISTORY to show the versions. |
| Time travel | Query an older version of a Delta table and a Snowflake table (Time Travel), and restore a table after a bad delete. |
| Roles and access | In Snowflake, create an analyst role that can only read one schema, grant it to a test user and prove the limits. |
Level 2: Building real features (10 tasks)
| Task | What to do |
|---|---|
| Medallion pipeline | Build bronze, silver and gold Delta tables from raw sales files: clean types, remove duplicates and create daily revenue by region. |
| Incremental loads with MERGE | Load a second day of changed and new rows with MERGE INTO so that the target table is upserted correctly. |
| Auto Loader or Snowpipe | Set up Databricks Auto Loader or Snowflake Snowpipe so that new files dropped in storage load automatically. |
| Star schema | Design and build a star schema (fact_sales with dim_date, dim_product, dim_customer) and draw the diagram. |
| Slowly changing dimension | Implement an SCD type 2 customer dimension that keeps the history when a customer changes city. |
| Semi-structured data | Load JSON order data into a Snowflake VARIANT column and query nested fields with path notation and FLATTEN. |
dbt models | Build your silver and gold transformations as dbt models with sources, refs and documentation, running on Snowflake or Databricks. |
| Data quality tests | Add dbt tests (not_null, unique, relationships, accepted_values) and a freshness check, and show a failing test caught. |
| Scheduling | Schedule your pipeline with Databricks Jobs, Snowflake Tasks or Airflow and show 3 successful scheduled runs. |
| Dashboard | Connect Power BI, Tableau Public or a Databricks SQL dashboard to your gold tables and build a sales dashboard. |
Level 3: Professional work and capstone (8 tasks)
| Task | What to do |
|---|---|
| Spark performance | Find a slow join, fix it with a broadcast join, partitioning or caching, and show the Spark UI before and after. |
| Cost tuning | Measure credits or DBUs used by your pipeline, then reduce them (smaller warehouse, auto-suspend, column pruning, clustering) and report the saving. |
| Change data capture | Use Snowflake streams and tasks or Delta change data feed to process only changed rows through your pipeline. |
| Governance | Use Unity Catalog or Snowflake masking policies to hide phone numbers from analysts and record the lineage of your gold table. |
| Machine learning on the lakehouse | Train a simple sales-forecast or churn model on your gold data in Databricks, track it with MLflow and write predictions back to a table. |
| Pipeline monitoring | Add row-count checks, failure alerts and a run log table to your pipeline and simulate a failed load. |
| Platform comparison | Write a comparison of Databricks and Snowflake for a mid-size Indian retailer: features, cost model, skills needed and your recommendation. |
| Capstone: end-to-end data platform | Ship a complete pipeline from raw files to dashboard with medallion layers, incremental loads, dbt tests, scheduling, governance and cost notes. Submit the repo, diagram and a 2-minute video. |