All projects
data
An ETL pipeline that fails loudly
Move data between systems, and stop bad data before it lands.
Why it matters
Silent data corruption is the worst failure mode in data work, because nobody notices for months. A pipeline that refuses to load bad rows is worth more than a faster one.
What you build
- Idempotent extraction and load
- Transformations with tests
- Quality checks that block a bad load
- A backfill you can rerun safely
If you want more
- Add incremental loading by watermark
- Add lineage tracking
Build it
Running it twice must not double the data.
- Rerun a historical window and get identical results
Material for the whole build
Shows you treat data correctness as a first-class requirement rather than an afterthought.
Get a plan built around projects like this