Data science supplies questioning, cleaning, visualization, and honest evaluation; AI/ML adds scalable pattern learning. Most real projects are data science workflows with ML steps, not pure model hacking.
Shared workflow
- Frame metric tied to business outcome
- Explore distributions and leakage
- Baseline with SQL or simple stats
- Iterate models with cross-validation
- Communicate uncertainty to stakeholders
Tools map
| Activity | Track |
|---|---|
| Arrays, vectorization | NumPy |
| Tables, joins, groupby | Pandas |
| Workflow, ethics, viz | Data science |
| Stats, optimization | SciPy |
Stakeholder chart
Data scientists explain recall vs precision trade-offs; engineers implement feature pipelines; PMs prioritize use cases by impact and risk.
Important interview questions and answers
- Q: DS without ML?
A: Yes—many insights come from aggregation and visualization alone. - Q: ML without DS discipline?
A: High risk of leakage, overfit metrics, and uninterpretable launches.
Self-check
- Map NumPy vs Pandas to activities.
- Why communicate uncertainty?
Tip: Many wins are clean data + SQL + visualization—ML is optional.
Interview prep
- DS without ML?
- Yes—exploration, SQL, and visualization often suffice.
- Shared workflow?
- Frame metrics, clean data, baseline, evaluate honestly, communicate uncertainty.