AI Data Engineer Feature Pipelines: Drafting a Lineage-Safe Transform

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-careers-ai-data-engineer-feature-pipeline-r9a4-adults

1. A data engineer is reviewing an AI-generated feature pipeline spec. Which task should the engineer verify manually rather than trusting the AI output?

   1. Confirming that join keys have been stable across all historical upstream data
   2. Generating a Mermaid diagram showing data lineage between tables
   3. Identifying inputs, transforms, and outputs for the pipeline
   4. Drafting SQL queries to monitor null rates in output columns

2. Which monitoring metric would an AI MOST likely be able to draft a query for without human intervention?

   1. Calculating the business impact of a feature being unavailable
   2. Detecting null values in the feature column over the past 24 hours
   3. Assessing whether a feature drift indicates a semantic schema change
   4. Determining whether the pipeline met its SLA during a system outage

3. A data engineer receives an AI-generated feature pipeline spec with a complete backfill plan. What should the engineer do before approving it for production?

   1. Run the backfill plan immediately in production to test it
   2. Delete the backfill plan to simplify the pipeline
   3. Accept the backfill plan as-is since the AI included one
   4. Request a worked example showing exactly how historical data would be recalculated

4. In feature pipeline terminology, what does 'lineage' refer to?

   1. The chain of command reporting structure within the data engineering team
   2. The chronological order of pipeline code commits in version control
   3. The genetic ancestry of the dataset used for training ML models
   4. The documented history of how data flows from source inputs through transforms to outputs

5. What artifact can an AI reliably generate to visualize data flow in a feature pipeline?

   1. A compiled Python executable that runs the entire pipeline
   2. A signed legal document agreeing to data quality standards
   3. A Mermaid diagram showing the lineage from source tables through transforms to output features
   4. A hardware diagram showing the physical servers that will host the pipeline

6. When reviewing an AI-generated pipeline spec, a data engineer notices a join key marked as 'unverified.' What is the correct interpretation?

   1. The join key is definitely incorrect and must be removed from the pipeline
   2. The join key should be left as-is because AI marked it correctly
   3. The AI has proven the join key is stable and production deployment can proceed
   4. The AI could not confirm the join key's stability and the engineer must verify it manually

7. What is a 'backfill' in the context of feature engineering pipelines?

   1. Filling missing data points with forward-filled values from the next valid record
   2. Recalculating historical feature values using new or corrected logic
   3. Creating a backup copy of the production pipeline code
   4. Running the pipeline in reverse order to validate output integrity

8. A data engineer discovers that an AI-generated feature pipeline spec includes a join between a customer table and a transaction table but does not explain how to handle customers with no transactions. What is the most appropriate response?

   1. Remove the join entirely to avoid complexity
   2. Request the AI revise the spec to include a left join or default value strategy
   3. Deploy the pipeline as-is since the AI must be correct
   4. Switch to an inner join to automatically filter out customers without transactions

9. Which statement best describes the division of labor between AI and data engineers in feature pipeline drafting?

   1. AI completely replaces the data engineer in the pipeline development process
   2. Data engineers draft the spec and AI validates it for errors
   3. AI drafts the specification while the data engineer owns correctness and validation
   4. AI and data engineers share equal responsibility for both drafting and validation

10. What type of data quality issue requires human judgment to detect rather than automated monitoring queries?

    1. The latency between data ingestion and feature availability
    2. The percentage of null values in a feature column over time
    3. Whether a join key has historically been stable across upstream source changes
    4. Whether feature values have drifted beyond a configured threshold

11. Why should feature pipeline specifications include a backfill plan?

    1. To allow the pipeline to run faster by skipping older data
    2. To ensure historical data can be correctly recomputed when transform logic changes
    3. To satisfy compliance auditors who require documentation of all code changes
    4. To enable automatic rollback if the pipeline fails during deployment

12. Which component is NOT typically part of an AI-generated feature pipeline specification?

    1. Monitoring queries for null rate, drift, and lateness
    2. A backfill plan for historical data recalculation
    3. A guarantee that join keys will never change in upstream systems
    4. Inputs, transforms, and outputs of the pipeline

13. What risk does an organization take if a data engineer approves an AI-generated pipeline spec without reviewing the backfill plan?

    1. Monitoring systems will automatically catch any backfill issues
    2. The pipeline will run faster since backfill logic is skipped
    3. The AI will automatically fix any backfill problems in production
    4. Historical feature values may become inconsistent when the pipeline is rerun

14. In feature engineering, what is the primary purpose of monitoring 'drift'?

    1. To detect when the statistical distribution of feature values changes significantly over time
    2. To monitor employee turnover in the data engineering team
    3. To measure how far behind schedule the pipeline execution runs
    4. To track whether the pipeline code has drifted from the original design

15. A junior data engineer asks why they can't just trust the AI-generated pipeline spec completely. What is the most accurate response?

    1. AI cannot verify join key stability or carry the on-call pager when things fail
    2. AI and humans have identical capabilities so trust is irrelevant
    3. The AI's specs are always wrong and must be completely rewritten
    4. AI is forbidden from generating pipeline specs by company policy