Home / Visual decks / A data science project, visually
Visual deck · 11 scenes · 27 animation beats
A data science project, visually
Decision, question, metric; the loop, not the pipeline; 80 % is data work; split before you touch anything; beat naive first; evaluate in the business currency; complexity is a loan; drift after deploy.
By Sitraka Forler · Lecturer, Durham Business SchoolUpdated 3 October 2026 About this site
Every data science course shows a pipeline: collect, clean, model, deploy. Real projects are a loop, and most of the time in the loop is spent on data and on the question, not on the model. This deck draws the loop and the places where projects actually fail.
It is built from the mistakes I see in student projects and in companies: no decision behind the question, a test set touched before the split, a model that never beat the naive baseline, and a dashboard nobody checks after launch. Each one gets a scene and a fix.
How to drive it
- Click or tap anywhere, or press Space or the right arrow: one press is one animation beat, so you can narrate between reveals.
- Left arrow or right click goes back one beat. Escape leaves the deck.
- The position is kept in the address bar (#scene.beat), so a link lands on the exact beat you were showing.
The 11 scenes
- 1Start from the decision someone has to make
- 2Turn the decision into a question and a metric
- 3The loop, not the pipeline
- 480 % of the work is data work
- 5Split before you touch: how leakage happens
- 6Beat the naive baseline first
- 7Evaluate in the business currency, not in accuracy
- 8Complexity is a loan with interest
- 9Drift: the world moves after deploy
- 10Four slides beat forty
- 11The checklist