Mu Sigma: Data Scientist

1. Fill the 5 L jug (5, 0).
2. Pour into the 3 L jug until it is full → (2, 3).
3. Empty the 3 L jug → (2, 0).
4. Pour the 2 L into the 3 L jug → (0, 2).
5. Fill the 5 L jug again → (5, 2).
6. Pour from the 5 L into the 3 L until it is full (1 L moves) → (4, 3). Done: the 5 L jug holds 4 L. - Explain your reasoning out loud. Interviewers grade the process (states, constraints), not just the answer. Since gcd(3, 5) = 1, any whole number of litres up to 5 is reachable.

• Convolutional layers apply small learned filters across the image, detecting local patterns (edges → textures → parts → objects). Weight sharing means few parameters and translation equivariance.
• Pooling downsamples for robustness; fully connected or global-pooling heads produce class scores.
• Training: backprop with cross-entropy; data augmentation; transfer learning from pre-trained models (ResNet, EfficientNet).
• Business use: shelf-image product recognition, defect detection, document classification.

• Clarify: which metric (revenue, units), which time frame, compared against what (last year, plan).
• Break it down: stores (same-store vs new or closed), categories, traffic × conversion × basket size, channels.
• Hypotheses: competitor opening, stock-outs, price changes, local events or weather, marketing changes.
• Data and tests: compare with control regions (difference-in-differences), check inventory logs, analyse promotion calendars.
• Recommend actions with the expected impact, and a monitoring plan.

Answer (STAR, 90 seconds):

• "In my churn project, the challenge was that 40% of the usage data was missing for older customers."
• "I investigated and found it was a logging change in 2023, so I created a flag feature and imputed using similar segments."
• "Result: AUC improved from 0.71 to 0.78, and the client used the top-decile list for a retention campaign that reduced churn by 6% in that group."

SELECT category, product, units
FROM (
  SELECT category, product, SUM(units) AS units,
         DENSE_RANK() OVER (PARTITION BY category ORDER BY SUM(units) DESC) AS rk
  FROM sales GROUP BY category, product
) t
WHERE rk = 2;

Answer (structure):

• Population about 1.4 crore; about 60% drink tea or coffee outside the home on a given day; about 50% of those choose tea, so about 42 lakh tea drinkers outside the home.
• About 1.5 cups each → about 63 lakh cups a day (plus office pantries). Sanity-check: about 50,000 tea stalls × about 120 cups ≈ 60 lakh cups. That matches.
• Clearly state the assumptions and adjust if the interviewer pushes back.