The Silence After 'Walk Me Through Your Notebook' Explains Everything

Interview Insights · Data Analytics

The Silence After 'Walk Me Through Your Notebook' Explains Everything

SECTION 01Why 'where do I even start' causes the pause

'Walk me through your notebook' is an open question, and open questions are hard when you have not decided on an opening. The notebook has fifty cells. Which one first? Should you mention the data source? The cleaning? The chart you are proud of? While you choose, the silence grows, and the interviewer reads it as uncertainty about your own work.

What the pause really reveals is that you prepared the notebook but not the answer. Those are different things. The notebook is a record of everything you did, in the order you did it. The answer is a short story, in the order a listener needs it. Good candidates know this, so they walk in with the first paragraph ready. They do not improvise the opening; they improvise only the follow-ups.

Interviewer: Walk me through your notebook.

Candidate: Okay ... so first I imported pandas, numpy
and seaborn. Then I loaded the CSV, it had about ...
some columns. Then I checked df.info() and df.head(),
and there were some nulls, so I handled those ...

(The interviewer still does not know what the
analysis found, or why it matters.)

Interviewer: Walk me through your notebook.

Candidate: This notebook looks at why food delivery
orders arrive late. The main finding is that late
deliveries are concentrated in the dinner peak and in
orders where the restaurant took long to prepare food,
not in long distances, as the team assumed.

I found that by comparing delay by hour, by distance
band and by preparation time. Happy to start with
whichever part interests you most.

The answer-first version takes about fifteen seconds and gives the interviewer three choices to explore. It also shows a result that challenged an assumption, which is one of the most interesting things an analysis can do. You do not need a surprising result, but if you have one, lead with it.

SECTION 02Build a summary cell that computes your key results

The easiest way to always have your first paragraph ready is to put it at the top of the notebook, as a summary cell. But do not type the numbers by hand. Hand-typed numbers drift out of date when you change the cleaning, and an interviewer who scrolls down and sees a different figure will notice.

Instead, write a small code cell near the end that computes the key results from your final data, and a markdown summary at the top that refers to them. Before the interview, run the notebook, check the summary matches, and memorise the two or three results. The code below uses a delivery dataset with order time, distance, preparation time and delivery time, and calculates the share of late orders by group.

import pandas as pd

d = pd.read_csv('data/deliveries_clean.csv', parse_dates=['order_time'])
d['late'] = d['delivery_minutes'] > d['promised_minutes']

d['hour_band'] = pd.cut(d['order_time'].dt.hour,
                        bins=[0, 11, 15, 18, 23],
                        labels=['morning', 'lunch', 'evening', 'dinner'],
                        include_lowest=True)
d['distance_band'] = pd.cut(d['distance_km'], bins=[0, 2, 4, 6, 50])
d['prep_band'] = pd.cut(d['prep_minutes'], bins=[0, 10, 20, 30, 120])

key = {
    'overall_late': d['late'].mean(),
    'by_hour': d.groupby('hour_band', observed=True)['late'].mean(),
    'by_distance': d.groupby('distance_band', observed=True)['late'].mean(),
    'by_prep': d.groupby('prep_band', observed=True)['late'].mean(),
}

print(f"Overall late share: {key['overall_late']:.1%}")
for name in ['by_hour', 'by_distance', 'by_prep']:
    print('\n' + name)
    print(key[name].map('{:.1%}'.format))

# Why do deliveries arrive late?

**Answer:** Late deliveries are concentrated in the
dinner peak and in orders with long restaurant
preparation time. Distance has a weaker link than the
team assumed.

**Evidence (see section 5, computed by key_results):**
1. Late share by hour band: highest at dinner.
2. Late share by distance band: small differences.
3. Late share by prep-time band: rises steadily.

**Method:** cleaned order logs, defined 'late' as
delivery time above promised time, compared late
share across groups.

**Limit:** association only; kitchen load is not
measured directly.

Notice that the summary cell describes patterns, and the exact percentages live in the computed output. In the interview you can quote the real numbers from your latest run, because you checked them that morning. That is the difference between a remembered number and an invented one.

SECTION 03Walking through a notebook with no screen at all

Sometimes there is no screen share. The interviewer says 'walk me through your notebook' in a phone screen, or in person with no laptop. This is where the silence is longest, because candidates who rely on scrolling have nothing to scroll. The answer-first structure works even better here, because it never needed the screen.

The trick is to describe the notebook as a few stages with one decision each, in plain language. Do not describe code, describe what happened to the data. After the opening paragraph, offer a short tour: 'There are four parts: cleaning, defining late, comparing groups and checking the result.' Then go through each part in two sentences. If the interviewer interrupts with a question, answer it and return to the next part. That stage list is your map when you cannot see the file.

"There are four parts.

First, cleaning. I removed cancelled orders, because a
cancelled order is not a late order, and rows with no
delivery time.

Second, defining late. I used delivery time above the
promised time, since promises differ by distance.

Third, comparing groups. I banded hour, distance and
preparation time and compared the late share.

Fourth, a check: within dinner orders only, longer
preparation was still linked to more late orders,
so it is not just the peak hour effect."

# Does prep time still matter within the dinner peak?
dinner = d[d['hour_band'] == 'dinner']

check = (dinner
         .groupby('prep_band', observed=True)
         .agg(orders=('late', 'size'),
              late_share=('late', 'mean')))

print(check)

# If late_share still rises with prep_band here,
# the prep-time link is not only a dinner-peak effect.

The fourth stage is what makes this walkthrough sound senior. Checking whether a pattern holds inside a subgroup is a simple idea, but freshers rarely do it. Mentioning it, even verbally, tells the interviewer you understand that two factors can be tangled together.

In Uncodemy mock interviews, we often ask candidates to do this verbal tour with their laptop closed. It quickly shows whether they know the story of their analysis or only the file.

SECTION 04Write your answer-first opening today

These three tabs help you write, test and rehearse the paragraph that ends the silence. Prepare one for every notebook you might be asked about.

ANSWER-FIRST OPENING (15-20 seconds)

"This notebook looks at ____________________.

The main finding is that ____________________,
[optional: not ____________, as was assumed].

I found that by ____________, ____________
and ____________.

Happy to start with whichever part interests you."

Read it aloud. If it takes more than 20 seconds,
cut words until it does not.

FOUR-STAGE VERBAL TOUR

Stage 1: ____________
  What happened to the data: ______________
  Decision: ______________________________

Stage 2: ____________
  What happened to the data: ______________
  Decision: ______________________________

Stage 3: ____________
  ...

Stage 4 (a check on your result): ________

Practise with the laptop closed.

BEFORE THE INTERVIEW

[ ] Restart and run all cells
[ ] Read the key_results output
[ ] Does the summary cell still describe it correctly?
[ ] Write the 2-3 numbers you will quote on a card
[ ] Check each number appears in a cell output
[ ] Remove any hand-typed number with no source

If a pattern changed after new cleaning,
update the summary and your opening paragraph.

SECTION 05Habits that make the first seconds easy

The silence disappears when the opening is a habit rather than a decision you make under pressure.

• Lead with the result — Your first sentence after 'walk me through' should contain what you found, not what you imported or loaded.
• Talk while the file opens — If there is a screen share, start your opening paragraph while the notebook loads. Waiting in silence for the file wastes your best seconds.
• Name the parts — Saying 'there are four parts' gives the interviewer a map and gives you a structure to return to after interruptions.
• Quote computed numbers only — Use only figures you saw in a fresh run of the notebook that day, and say where they come from if asked.
• Hand back control — End the opening by offering a choice of where to go next. It turns a monologue into a conversation.

SECTION 06Interview Q&A — Notebook question silence

SECTION 07Test yourself — Notebook question silence quiz

SECTION 08Frequently asked questions

SECTION 09Related reads