Building AI Data Pipelines

8 Interactive Sessions

Short, interactive sessions — watch it work, steer it, then build it yourself. Go deeper anytime with the full code walkthrough.

1. 1

   The pipeline runs green — and the data still lied

   See why a data analyst's real job is noticing when a number is lying, and why that starts with landing the source faithfully.

2. 2

   Trust starts at typing

   Turn the faithful-but-useless raw copy into a typed, trustworthy staging table — and see why idempotency is what makes a re-run safe.

3. 3

   Catch the bad row — don't drop it, don't ship it

   Test staging with predicates that should return zero rows, and quarantine the ones that fail — held for review, not deleted and not passed through.

4. 4

   The table the business trusts

   Publish the daily metrics mart from clean staging only — and see, in a single number, the quality gate protecting a headline metric.

5. 5

   The failure no row count can see

   Detect a distribution that shifted while every count stayed fine — using PSI against a saved baseline, with a stable control proving the detector isn't crying wolf.

6. 6

   The pipeline that runs perfectly on yesterday's data

   Catch silent staleness — a pipeline that runs green while reading data that stopped arriving — with a freshness SLA, and emit a run manifest for lineage.

7. 7

   The model drafts; you own the merge

   Use a model to draft the repetitive 80% of pipeline work — transforms, tests, explanations — while keeping human judgement as the gate on every line that ships.

8. 8

   Which signal wakes a human at 3am?

   Turn the pipeline's checks into a paging policy people trust — page only for the failures that hurt if missed, WARN for the rest, and close the loop back to the baseline.

What you walk away with