Pulls the data behind a product question, builds the analysis or model that answers it, and reports what the numbers do and do not support.
Build your own roadmap from your CVEvery analysis, model and notebook you hand over is Python; there is no other language on the team.
The data you model lives in a warehouse; you pull your own training sets, nobody pulls them for you.
Cleaning, joining and reshaping the table is most of the day, and pandas is where it happens.
A model is a claim about data; you must know when the claim is supported and when it is noise.
You will be asked which model fits the problem and why, before you are asked to train it.
The baseline for every tabular problem starts here; you ship a working pipeline before anything fancier.
Half the questions you get have no model in the answer; you need to find the pattern and explain it.
A finding nobody can see is not a finding; the chart is the deliverable as often as the model.
Most of the lift comes from what you feed the model, not the model; you build and test the features.
Accuracy on the training set convinces nobody; you pick the metric that matches the business cost.
Notebooks and pipelines are reviewed in pull requests; you commit small and often.
Every experiment readout ends with a p-value someone will act on; you must know what it does and does not say.
Product decisions here are settled by experiments; you design them, size them and read them.