Universal topic learning · Practical workbook

Data Science Prompt Pack

Tick the checklist and type your answers; they stay in this browser. Print or save as PDF with the button below. The review guide prints on a separate page.

Example prompts for each step of a data science project, and a worksheet to build your own reusable versions. Use your approved AI assistant with fictional or approved data only.

Practise with fictional information. Use your approved procedures for real work.

Before you begin

Give the role, context, task, constraints and output format.
Share the schema (column names, types and a few made-up rows), not the real data.
Ask for a plan before any code, then one step at a time.
Ask the AI to state its assumptions and what it cannot know.
Ask it to explain each line of code and to flag anything it is unsure of.
Run the code on a small sample with a known answer before trusting it.
Check every column, function and figure in the answer against your data.
Save the prompt, the model version and the date with the result.

Your planning worksheet

Type your answers below. They are saved in this browser only; download or print them when you finish.

Profiling prompt (schema only)

Cleaning prompt and the change table you expect

SQL prompt and the row-count check

Chart prompt and the question it answers

Modelling plan prompt and the leakage check

Evaluation prompt and the costs of each mistake

Explain-results prompt and the audience

Debugging prompt (error text, no data or secrets)

Template owner, version and review date

Practice scenario

Use the fictional Kopi Kita file kk_orders.csv (order_id, member_id, outlet_name, order_date, sales_channel, item_qty, total_rm). Write your own prompt for each worksheet line, then compare it with the example prompts in the review guide.

Review guide

Profiling: "You are helping a data analyst. Here is the schema of a fictional orders file: [columns and types]. Write Python code with a dataframe library that reports row count, types, missing %, unique values, min and max, and duplicate order_id. Explain each step. Do not change the data."
Cleaning: "Propose cleaning steps for these issues: dates in two formats, outlet names spelt three ways, total_rm stored as text like RM1,200. Show a change table (old → new) for me to approve before any code. Mark gaps as missing; do not guess."
SQL: "Tables: kk_orders(order_id, outlet_id, order_date, total_rm), kk_outlets(outlet_id, outlet_name, region). Write a read-only query for total sales and order count per outlet per month. Add checks for row counts before and after the join and for duplicate outlet_id."
Chart: "Weekly sales for five outlets over 12 weeks. Which outlets are growing? Recommend one chart and why, write the plotting code, label axes with units, start bars at zero and give a title that states the finding."
Modelling plan: "Plan a churn model: 37,000 members, churn = no purchase for 60 days after the snapshot date. List target, features, a time-based split, a simple baseline, two candidate models and metrics. Flag any feature that could leak the answer. No code yet."
Evaluation: "A wasted voucher costs RM5 and a lost member about RM120. About 15% leave. Which metrics should I report, how do I pick the threshold from these costs, and what baseline should I compare with? Show the arithmetic."
Explaining: "Turn these results into a one-page summary for the head of finance: answer first, three supporting points, the main limit, a recommendation with an owner and the decision needed. Use only these figures and explain any technical term."
Debugging: "My code fails with this error: [error text]. The column came from a file with amounts like RM1,200. Explain the cause and the smallest fix, and give one check to prove it worked. I removed paths, data and passwords."

Primary references

NIST AI Risk Management Framework

Project-authored exercises. References support the underlying principles; this workbook does not prescribe company procedures.

Practice scenario: review guide

Profiling: "You are helping a data analyst. Here is the schema of a fictional orders file: [columns and types]. Write Python code with a dataframe library that reports row count, types, missing %, unique values, min and max, and duplicate order_id. Explain each step. Do not change the data."
Cleaning: "Propose cleaning steps for these issues: dates in two formats, outlet names spelt three ways, total_rm stored as text like RM1,200. Show a change table (old → new) for me to approve before any code. Mark gaps as missing; do not guess."
SQL: "Tables: kk_orders(order_id, outlet_id, order_date, total_rm), kk_outlets(outlet_id, outlet_name, region). Write a read-only query for total sales and order count per outlet per month. Add checks for row counts before and after the join and for duplicate outlet_id."
Chart: "Weekly sales for five outlets over 12 weeks. Which outlets are growing? Recommend one chart and why, write the plotting code, label axes with units, start bars at zero and give a title that states the finding."
Modelling plan: "Plan a churn model: 37,000 members, churn = no purchase for 60 days after the snapshot date. List target, features, a time-based split, a simple baseline, two candidate models and metrics. Flag any feature that could leak the answer. No code yet."
Evaluation: "A wasted voucher costs RM5 and a lost member about RM120. About 15% leave. Which metrics should I report, how do I pick the threshold from these costs, and what baseline should I compare with? Show the arithmetic."
Explaining: "Turn these results into a one-page summary for the head of finance: answer first, three supporting points, the main limit, a recommendation with an owner and the decision needed. Use only these figures and explain any technical term."
Debugging: "My code fails with this error: [error text]. The column came from a file with amounts like RM1,200. Explain the cause and the smallest fix, and give one check to prove it worked. I removed paths, data and passwords."