Prepare for and practice interview questions from Itw.

Itw Interview Questions

Itw Interview Guides

Machine Learning

Which model would perform better and why?

Booking Regression

How do we deal with the missing square footage data to construct our model

Missing Housing Data

How would we build a model to detect fraud and text customers to approve or deny fraudulent transactions

Bank Fraud Model

Rebalancing outcome probabilities for a classifier on imbalanced data.

Rebalance Probabilities

Lasso vs Ridge

Data Structures & Algorithms

Given two sorted lists, write a function to merge them into one sorted list.

Merge Sorted Lists

Given an integer N, write a function that returns all of the prime numbers up to N

Prime to N

Write a program in python to determine the TF values for each term of a document

Term Frequency

This problem involves identifying duplicate numbers in a list of integers. The function should return a list of the duplicate numbers.

Find Duplicate Numbers in a List

Write a function `sort_lists` to combine sorted integer lists into one sorted list without using libraries or `sort`/`sorted`

Merge N Sorted Lists

Statistics

What are the assumptions of linear regression?

Assumptions of Linear Regression

Explain the difference between covariance and correlation and provide an example

Covariance vs Correlation

How do you detect and handle correlation between variables in linear regression, and what happens if you ignore it

Correlation in Regression

What does it mean to "bootstrap" a data set?

Bootstrapping Samples

What are the regression coefficients when regressing Y on X versus X on Y, given Y = X plus normal white noise?

Regress Y on X

Analytics

How would you present the performance of each subscription to an executive?

Analyzing Churn Behavior

Let’s say that you're in charge of an e-commerce D2C business that sells socks. What business health metrics would you care?

D2C Socks e-Commerce

Which metric should we use for late orders if the distribution is right-skewed and what does that tell us?

Late Orders

Let's say you work at Facebook and you're analyzing churn on the platform.

Retention Rate Disparity

Describing a real-world data cleaning and organization project

Data Cleaning Experiences

Supply Chain

Optimizing Supply Chain Efficiency

How would you decide on a metric and approach for worker allocation across an uneven production line?

Worker Distribution Dilemma

Which three metrics would you prioritize to evaluate supply chain efficiency at Niagara Bottling?

Supply Chain Efficiency Metrics

How would you handle a sole supplier demanding a steep price increase when resourcing isn’t an option?

Product Procurement Predicament

Tell me about a data project that didn’t go the way you expected. What did you set out to do, what surprised you, and how did you handle it?

Describing a data project and its challenges

Hurdles In Data Projects

Behavioral

Write a SQL query to select the 2nd highest salary in the engineering department.

Note: If more than one person shares the highest salary, the query should select the next highest salary.

Example:

Input:

<code>employees</code> table

<table>
<thead>
<tr>
<th>Column</th>
<th>Type</th>
</tr>
</thead>

<tbody>
<tr>
<td><code>id</code></td>
<td>INTEGER</td>
</tr>

<tr>
<td><code>first_name</code></td>
<td>VARCHAR</td>
</tr>

<tr>
<td><code>last_name</code></td>
<td>VARCHAR</td>
</tr>

<tr>
<td><code>salary</code></td>
<td>INTEGER</td>
</tr>

<tr>
<td><code>department_id</code></td>
<td>INTEGER</td>
</tr>
</tbody>
</table>
<code>departments</code> table

<table>
<thead>
<tr>
<th>Column</th>
<th>Type</th>
</tr>
</thead>

<tbody>
<tr>
<td><code>id</code></td>
<td>INTEGER</td>
</tr>

<tr>
<td><code>name</code></td>
<td>VARCHAR</td>
</tr>
</tbody>
</table>
Output:

<table>
<thead>
<tr>
<th>Column</th>
<th>Type</th>
</tr>
</thead>

<tbody>
<tr>
<td><code>salary</code></td>
<td>INTEGER</td>
</tr>
</tbody>
</table>

Select the 2nd highest salary in the engineering department

2nd Highest Salary

Given two sorted lists, write a function to merge them into one sorted list.

Bonus: What’s the time complexity?

Example:

Input:

<pre tabindex="0" class="chroma"><code>list1 = [1,2,5]
list2 = [2,4,6]
</code></pre>

Output:

<pre tabindex="0" class="chroma"><code>def merge_list(list1,list2) -&gt; [1,2,2,4,5,6]
</code></pre>

You have two sorted lists, one with m elements and one with n elements. What would be the time complexity of merging the two lists via merge sort?

You have two sorted lists, one with m elements and one with n elements. What would be the time complexity of merging the two lists via merge sort? 

Merging Lists Complexity

Given an integer <code>N</code>, write a function that returns a list of all of the prime numbers up to <code>N</code>.

Note: Return an empty list there are no prime numbers less than or equal to <code>N</code>.

Example:

Input:

<pre tabindex="0" class="chroma"><code>N = 3
</code></pre>

Output:

<pre tabindex="0" class="chroma"><code>def prime_numbers(N) -&gt; [2,3]
</code></pre>

Let’s say we’re comparing two machine learning algorithms. In which case would you use a bagging algorithm versus a boosting algorithm? 

Give an example of the tradeoffs between the two.

In machine learning, when would you use a bagging algorithm over a boosting algorithm?

In machine learning, when would you use a bagging algorithm over a boosting algorithm?

Bagging vs. Boosting

In which case would you use a bagging algorithm versus a boosting algorithm

Bagging vs Boosting

Let’s say that you work at a bank that wants to build a model to detect fraud on the platform.

The bank wants to implement a text messaging service in addition that will text customers when the model detects a fraudulent transaction in order for the customer to approve or deny the transaction with a text response.

How would we build this model?

Let’s say that you work at a bank that wants to build a model to detect fraud on the platform.

The bank wants to implement a text messaging service that will text customers when the model detects a fraudulent transaction in order for the customer to approve or deny the transaction with a text response.

Which statement is true?

Let's say that you work at a bank that wants to build a model to detect fraud on the platform.

The bank wants to implement a text messaging service that will text customers when the model detects a fraudulent transaction in order for the customer to approve or deny the transaction with a text response.

Which statement is true?

Let’s say we want to build a model to predict booking prices on Airbnb.

Between linear regression and random forest regression, which model would perform better and why?

What are the two key concepts of random forest regression?

What are the two key concepts of random forest regression?

Let’s say you work at Allstate. Allstate is running <code>N</code> online ads right now. The table <code>ads</code> contains all those ads, ranked by popularity via the <code>id</code> column (e.g., the entry with <code>id = 1</code> is the most popular, etc.).

Create a subquery or common table expression named <code>top_ads</code> containing the top 3 ads (by popularity) and return the number of rows that would result from the following operations

<ol>
<li><code>ads INNER JOIN top_ads</code></li>
<li><code>ads LEFT JOIN top_ads</code></li>
<li><code>ads RIGHT JOIN top_ads</code></li>
<li><code>ads CROSS JOIN top_ads</code></li>
</ol>

Note: Please make the <code>join_type</code> column in your output have the values <code>inner_join</code>, <code>left_join</code>, etc. for each of their respective join types

Note: Please return only one query with each number in a different row

Example:

Input:

<table>
<thead>
<tr>
<th>Column</th>
<th>Type</th>
</tr>
</thead>

<tbody>
<tr>
<td><code>id</code></td>
<td>INTEGER</td>
</tr>

<tr>
<td><code>name</code></td>
<td>VARCHAR</td>
</tr>
</tbody>
</table>
Output:

<table>
<thead>
<tr>
<th>Column</th>
<th>Type</th>
</tr>
</thead>

<tbody>
<tr>
<td><code>join_type</code></td>
<td>VARCHAR</td>
</tr>

<tr>
<td><code>number_of_rows</code></td>
<td>INTEGER</td>
</tr>
</tbody>
</table>

Create top_ads with the top 3 ads and return the row counts for inner, left, right, and cross joins with ads

Size of Joins

You’re given a string that may contain the characters <code>{</code>, <code>}</code>, <code>[</code>, <code>]</code>, <code>(</code>, and <code>)</code>.

Task: Verify that the string is balanced. A balanced string is one where every opening character, <code>{</code>, <code>[</code>, or <code>(</code>, has a corresponding closing character, <code>}</code>, <code>]</code>, or <code>)</code>.

Write a function called <code>is_balanced(string: str) -&gt; bool</code> which verifies the balance of a string.

Example:

<pre tabindex="0" class="chroma"><code>is_balanced(&#39;(())[]{}&#39;) -&gt; True
</code></pre>

<pre tabindex="0" class="chroma"><code>is_balanced(&#39;{([(){}])()}&#39;) -&gt; True
</code></pre>

<pre tabindex="0" class="chroma"><code>is_balanced(&#39;{}[]())&#39;) -&gt; False
</code></pre>

<hr/>

Write a function that tests whether a string of brackets is balanced.

The Brackets Problem

You work as a data engineer at a global financial services company. The current pipeline ingests millions of daily credit card transactions via nightly batch jobs. The business now requires real-time processing and analysis for each transaction to support instant fraud detection and real-time reporting.

How would you redesign the ingestion and processing architecture to enable reliable, scalable, and consistent real-time streaming of transaction data?

Draft a design proposal outlining your approach.

Redesign batch credit card processing to enable real-time streaming for fraud detection and reporting.

Real-Time Transaction Streaming

Data Pipelines

Given a 2D terrain represented by an array of non-negative integers, where each integer represents the height of a terrain level at that index, implement an algorithm to calculate the total amount of rainwater that can be trapped in this terrain.
Consider that rainwater can only be trapped between two terrain levels with higher heights, and the trapped water cannot flow out through the edges.

The algorithm should have an optimal time complexity of O(n) and a space complexity of O(n).
 Provide a detailed explanation of the algorithm and its implementation in Python.

Example:

<table>
<thead>
<tr>
<th>Index</th>
<th>0</th>
<th>1</th>
<th>2</th>
<th>3</th>
<th>4</th>
</tr>
</thead>

<tbody>
<tr>
<td>Terrain Level</td>
<td>3</td>
<td>0</td>
<td>2</td>
<td>0</td>
<td>4</td>
</tr>

<tr>
<td>Trapped Water</td>
<td>0</td>
<td>3</td>
<td>1</td>
<td>3</td>
<td>0</td>
</tr>
</tbody>
</table>
In this table, the ‘Index’ row represents the position of the terrain levels, the ‘Terrain Level’ row shows the height of the terrain at each index, and the ‘Trapped Water’ row illustrates the trapped rainwater between the terrain levels. The total trapped rainwater in this example is 7 units (3 + 1 + 3).

Input:

<pre tabindex="0" class="chroma"><code>terrainLevels = [3, 0, 2, 0, 4]
</code></pre>

Output:

<pre tabindex="0" class="chroma"><code>7
</code></pre>

In this example, the terrain is represented by the <code>terrainLevels</code> array. The algorithm should calculate the total trapped rainwater, which is 7 units in this case. The trapped rainwater is formed between the heights of 3, 0, 2, 0, and 4.

Given a 2D terrain array, calculate the total amount of trapped rainwater with O(n) time and O(n) space

Level Of Rain Water In 2D Terrain

What is the difference between covariance and correlation? Provide an example.

How does random forest generate the forest? Additionally, why would we use it over other algorithms such as logistic regression?

What happens when you average the output on multiple decision trees?

What happens when you average the output on multiple decision trees?

Average Trees

How does random forest generate the forest and why use it over logistic regression

Random Forest Explanation

Given a <code>ListNode</code> representing the head of a linked list, write a function <code>is_cyclic</code> which returns <code>True</code> if a linked list is cyclic and <code>False</code> if it is not. A linked list is cyclic when there is no tail to the linked list, and the supposed tail is attached to another node inside the list itself, creating a cycle.

A <code>ListNode</code> is defined as:

<pre tabindex="0" class="chroma"><code>class ListNode:
	value: ListNode = None
 next: ListNode = None
</code></pre>

Example:

Input:

<img src="/s3/questions/1a65651d-a483-4481-b073-2738d65ab9d7.png" alt="image"/>

Output:

<pre tabindex="0" class="chroma"><code>def is_cyclic(head) -&gt; True
</code></pre>

Cyclic Detection

Write a Python function called <code>longest_increasing_subsequence</code> that takes a list of integers as input and returns as output the length of the longest increasing subsequence in the list.

Note: An increasing subsequence is a subset of a list where the elements are in strictly increasing order. The subsequence does not have to be contiguous or unique.

<h3>Example:</h3>

Input:

<pre tabindex="0" class="chroma"><code>nums = [10, 9, 2, 5, 3, 3, 7, 101, 18]
</code></pre>

Output:

<pre tabindex="0" class="chroma"><code>4
</code></pre>

Explanation:

The longest increasing subsequence is <code>[2, 3, 7, 101]</code>, for a total length of 4.

Find the longest increasing subsequence in a list of integers.

Longest Increasing Subsequence

We want to build a model to predict housing prices in the city of Seattle. We’ve scraped 100K sold listings over the past three years but found that around 20% of the listings are missing square footage data.

How do we deal with the missing data to construct our model?

We want to build a model to predict housing prices in the city of Seattle. We’ve scraped 100K sold listings over the past three years, but have discovered that around 20% of the listings are missing square footage data.

How would you approach dealing with this missing data, in order to construct the most useful predictive model possible?

We want to build a model to predict housing prices in the city of Seattle. We’ve scraped 100K sold listings over the past three years, but have discovered that around 20% of the listings are missing square footage data.

How would you approach dealing with this missing data, in order to construct the most useful predictive model possible?



You’re AI engineer working on an energy forecasting system at Siemens that helps utilities predict short-term electricity demand to balance grid load and reduce outages. You train a recurrent neural network on long historical time series (weather, calendar effects, local demand signals) to forecast energy demand.

Early in training, the model appears to learn normally, but after several epochs the loss becomes highly unstable, oscillating wildly before turning into NaNs. At the same time, training slows dramatically and the model fails to recover even after restarting from recent checkpoints.

You’ve verified that the data pipeline is correct and labels are well-aligned with inputs. No changes were made to the dataset, but increasing the sequence length or learning rate makes the problem significantly worse.

How would you investigate the cause of this training instability? What signals would you look for during debugging, and how would those findings guide the changes you make to the model or training setup?

How would you investigate the cause of this training instability?

Training Instability in Neural Networks

Say you are given a text document in the form of a string with the following sentences:

Example:

Input:

<pre tabindex="0" class="chroma"><code>document = &#34;I have a nice car with a nice tires&#34;
</code></pre>

Output:

<pre tabindex="0" class="chroma"><code>
{
&#34;I&#34;:0.11,
&#34;have&#34;:0.11,
&#34;a&#34;:0.22,
&#34;nice&#34;:0.22,
&#34;car&#34;: 0.11,
&#34;with&#34;:0.11,
&#34;tires&#34;:0.11
}
</code></pre>

Write a program in python to determine the TF (<code>term_frequency</code>) values for each term of this document.

Note: Round the term frequency to 2 decimal points.

Given a list of integers, identify all the duplicate values in the list. Assume that the list can contain both positive and negative numbers, and the order of the list does not matter. A number is considered a duplicate if it appears more than once in the list. Return a list of the duplicate numbers.

Example 1:

Input:

<pre tabindex="0" class="chroma"><code>nums = [1, 2, 3, 1, 2, 3]
</code></pre>

Output:

<pre tabindex="0" class="chroma"><code>find_duplicates(nums) -&gt; [1, 2, 3]
</code></pre>

The numbers 1, 2, and 3 all appear more than once in the list, so they are considered duplicates.

Example 2:

Input:

<pre tabindex="0" class="chroma"><code>nums = [1, -1, 2, 3, 3, -1]
</code></pre>

Output:

<pre tabindex="0" class="chroma"><code>find_duplicates(nums) -&gt; [-1, 3]
</code></pre>

The numbers -1 and 3 both appear more than once in the list, so they are considered duplicates. Note that the order of the output does not matter.

Example 3:

Input:

<pre tabindex="0" class="chroma"><code>nums = [1, 2, 3, 4, 5]
</code></pre>

Output:

<pre tabindex="0" class="chroma"><code>find_duplicates(nums) -&gt; []
</code></pre>

None of the numbers in the list appear more than once, so there are no duplicates.

What’s the difference between Lasso and Ridge Regression?

Under what circumstances is it recommended to use Lasso regression over Ridge regression?

Under what circumstances is it recommended to use Lasso regression over Ridge regression?

You work as a machine learning engineer at a fintech company. You are tasked with designing a system to standardize feature storage and retrieval for all production ML models that predict credit risk. The system must ensure consistent access to customer transaction data, account histories, and real-time financial events, while supporting rapid model deployment using AWS SageMaker.

How would you architect a feature store and integrate it into the end-to-end ML pipeline on AWS SageMaker?

Design a feature store for credit risk ML models and integrate it with SageMaker.

Feature Store Integration

ML Ops

What are the assumptions of linear regression?

Which assumption of the residuals of the standard linear regression model can not be overcome by increasing the sample size?

Regression assumptions

Let’s say you work as a data architect at Facebook.

Facebook is trying to transition their product from just doing likes on posts and comments to adding the multi-reaction under the like button.

How would you approach this problem? What would the database modifications look like?

How would you approach adding multi-reactions under the like button and what database modifications would you make

Multi-Reaction

Data Modeling

Let’s say you work as an LLM engineer at an enterprise software company. You are tasked with designing a retrieval-augmented generation (RAG) system to ensure that the language model only provides answers based on a specific set of ingested documents, and never responds using its original training data.

How would you architect this system?

Design a RAG system that restricts LLM responses exclusively to ingested data, preventing use of pre-trained knowledge.

RAG Strict Source Control

AI & Agentic Systems

Given a list of sorted integer lists, write a function <code>sort_lists</code> to create a combined list while maintaining sorted order without importing any libraries or using the <code>&#39;sort&#39;</code> or <code>&#39;sorted&#39;</code> functions in Python.

Example:

Input:

<pre tabindex="0" class="chroma"><code>lists = [
[1,2,3,4,5,6],
[2,5,7,8],
[3,9,10,12],
[0,1,2,8]
]
</code></pre>

Output:

<pre tabindex="0" class="chroma"><code>def sort_lists(lists) -&gt; [0,1,1,2,2,3,3,4,5,5,6,7,8,8,9,10,12]
</code></pre>

What are the key differences between classification models and regression models?

What is the MOST important difference between regression and classification models?

What is the MOST important difference between regression and classification models?

Classification vs Regression

Explain the key differences between classification models and regression models

Classification and Regression

Let’s say we’re managing a production line for assembling electric scooters, with 10 stations where each has a different average processing time per unit (for example, Station 1 takes 5 minutes, Station 2 takes 3 minutes, and so on).

How would you determine the right metric to use for balancing worker allocation across the stations, and how would you use it to figure out how many workers each station needs?

You work as a software engineer at Credit Karma. The team is tasked with building a backend API that retrieves and serves relevant financial product offers, such as credit cards, personal loans, and auto loans, based on user profiles.

When a user logs in, the API must fetch all available offers from multiple partner services, filter them using factors like credit score, and return only the most relevant auto loan offers if the user expresses interest.

How would you architect the API endpoints and backend logic to support this functionality?

Design a backend API for serving filtered, relevant financial offers based on user profiles.

Offer Matching API Design

Web Engineering

Imagine you are asked to build a machine learning model to decide new loan approvals for a
financial firm. You ask the data department in the company for a subset of data to get started
working on the problem. The data includes different features about applicants such as age,
occupation, zip code, height, number of children, favorite color, etc. You decide to build
multiple machine learning models to test out different ideas before settling on the best one.

How would you explain the bias-variance tradeoff with regards to
building and choosing a model to use?

Bias vs. Variance Tradeoff

How would you handle the data preparation for building a machine learning model using imbalanced data?

Addressing imbalanced data in machine learning through carefully prepared techniques.

Calculate Moving Average	SQL	Easy
Predict Customer Churn	Machine Learning	Medium
A/B Test Significance	Statistics	Medium
Optimize Query Performance	SQL	Hard
Feature Importance Analysis	Machine Learning	Medium
Clean Missing Data	Python	Easy
Neural Network Architecture	Deep Learning	Hard
Calculate Cohort Retention	SQL	Medium
Bayesian Probability	Statistics	Easy
Recommend Similar Products	Machine Learning	Hard

Itw Interview Questions

Itw Interview Guides

Itw Interview Questions

Challenge

Discussion & Interview Experiences

Discussion & Interview Experiences