Prepare for and practice interview questions from Pratt & Whitney across topics like Statistics, SQL, Machine Learning and more.

Pratt & Whitney Interview Questions

Pratt & Whitney Interview Guides

Machine Learning

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

Bank Fraud Model

Lasso vs Ridge

Rebalancing outcome probabilities for a classifier on imbalanced data.

Rebalance Probabilities

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

Bagging vs Boosting

What kind of model did the co-worker develop, how would you measure the difference between the two credit risk models within a timeframe, and what metrics would you track to measure the success of the new model

Loan Model

Data Structures & Algorithms

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

Merge N Sorted Lists

Implementing a priority queue used linked lists.

Priority Queue Using Linked List

Implement a fixed-length array with addition, deletion, and search operations.

Fixed-Length Arrays: Deletion

This question requires the implementation of the Fibonacci sequence using three different methods: recursively, iteratively, and using memoization.

Implementing the Fibonacci Sequence in Three Different Methods

Write a class 'SinglyLinkedList' that implements the interface required for a Singly Linked List.

Singly Linked List

Behavioral

Describing a data project and its challenges

Hurdles In Data Projects

Strategically resolving misaligned expectations with stakeholders for a successful project outcome

Stakeholder Communication

How to present complex data insights with clarity and adaptability tailored to a specific audience

Presentations and Insights

How would you negotiate and resolve disagreements when a client rejects your proposed solution?

Client Solution Pushback

How would you answer when an Interviewer asks why you applied to their company?

Why Do You Want to Work With Us

Analytics

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

Analyzing Churn Behavior

How would you approach improving the quality of airline data?

Addressing Data Quality Issues

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

You’re tasked with analyzing data from multiple sources, such as payment transactions, user behavior, and fraud detection logs. How would you approach solving a data analytics problem involving these diverse datasets? What steps would you take to clean, combine, and extract meaningful insights that could improve the system's performance?

Analyzing Multiple Data Sources

Business Case

How would you evaluate and choose between a fast, simple model and a slower, more accurate one for product recommendations?

Deciding Between Solutions

As a data scientist at a mortgage bank, how would you approach building a predictive model for loan default risk?

Risk Model for a Mortgage Bank

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?

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’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

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

Explain the difference between covariance and correlation and provide an example

Covariance vs Correlation

Statistics

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

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?

Priority queues are an abstract data structure responsible for allowing enqueuing items with an attached priority. While typically implemented with a heap, implement a priority queue using a linked list.

The Priority Queue implementation should support the following operations:

<ul>
<li><code>insert</code>(element, priority): This operation should be able to insert an element into the Priority Queue, along with its corresponding priority.</li>
<li><code>delete()</code>: This operation should remove and return the element with the highest priority. If multiple elements share the same highest priority, the element first enqueued should be returned. In the case that the queue is empty, return None.</li>
<li><code>peek()</code>: This operation should return the element with the highest priority without removing it from the Priority Queue. Again, in the case of equal highest priorities, the element first enqueued should be returned. In the case that the queue is empty, return None.</li>
</ul>

Note: Smaller priority values imply that they have higher priority.

<h3>Input</h3>

<pre tabindex="0" class="chroma"><code>queue.insert(2, 1)
queue.insert(3, 3)
</code></pre>

<h3>Output:</h3>

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

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>

Every night between 7 pm and midnight, two computing jobs from two different sources are randomly started with each one lasting an hour.

Unfortunately, when the jobs simultaneously run, they cause a failure in some of the company’s other nightly jobs, resulting in downtime for the company that costs $1000. 

The CEO, who has enough time today to hear one word, needs a single number representing the annual (365 days) cost of this problem.

Note: Write a function to simulate this problem and output an estimated cost 

Bonus - How would you solve this using probability?

Example:

Input:

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

Output:

<pre tabindex="0" class="chroma"><code>simulate_overlap(n) -&gt; 0.4
</code></pre>

Find how much overlapping jobs are costing the company

Nightly Job

Probability

Let’s say you work for a bank that gives out personal loans. Your co-worker develops a model that takes in customer inputs and returns if a loan should be given or not.

<ol>
<li>What kind of model did the co-worker develop? 
</li>
<li>Another co-worker thinks they have developed a better model to predict defaults on the loans. Given that personal loans are monthly installments of payments, how would you measure the difference between the two credit risk models within a timeframe? 
</li>
<li>What metrics would you track to measure the success of the new model?</li>
</ol>

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

The Fibonacci sequence is a series of numbers in which each number is the sum of the two preceding ones, usually starting with 0 and 1. It is often used in algorithm examples, and is defined by the following formula: F(n) = F(n-1) + F(n-2), with F(0) = 0 and F(1) = 1.

Your task is to implement the Fibonacci algorithm in three different methods:
1. Recursively
2. Iteratively
3. Using Memoization

Example 1:

Input:

<pre tabindex="0" class="chroma"><code>n = 5
</code></pre>

Output:

<pre tabindex="0" class="chroma"><code>fibonacci(n) -&gt; 5
</code></pre>

Example 2:

Input:

<pre tabindex="0" class="chroma"><code>n = 10
</code></pre>

Output:

<pre tabindex="0" class="chroma"><code>fibonacci(n) -&gt; 55
</code></pre>

The Fibonacci sequence starts as follows: 0, 1, 1, 2, 3, 5, 8, 13, 21, 34, 55…

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

Let’s say you’re a senior data analyst and your team is tasked with addressing data quality issues for an airline company. Customer records, booking details, and flight schedules often have missing fields, duplicate entries, or incorrect data.

How would you approach improving the quality of this data and ensuring it remains clean and reliable over time?

You have been tasked with designing three classes: <code>text_editor</code>, <code>moving_text_editor</code>, and <code>smart_text_editor</code>. These classes are to be created with specific functionalities as defined below:

<ol>
<li><code>text_editor</code> class:

<ul>
<li><code>write_line(string:str)</code>: A method which appends a given string to the end of the existing string.</li>
<li><code>delete_line(char_num : int)</code>: A method which deletes <code>char_num</code> number of characters from the existing string, starting from the end. If there are no characters left, the method should do nothing.</li>
<li><code>special_operation()</code>: A method which currently does nothing.</li>
<li><code>get_notes()</code>: A method which returns the internal string.</li>
</ul></li>

<li><code>moving_text_editor</code> class:

<ul>
<li>This class extends <code>text_editor</code>. The <code>special_operation()</code> method is overridden. Initially, the cursor will be at the end of the current string. If <code>special_operation()</code> is called, it moves the cursor to the beginning of the string, any additional appends will be appended to the beginning of the string instead. Calling <code>special_operation()</code> again reverses the cursor operation.</li>
</ul></li>

<li><code>smart_text_editor</code> class:

<ul>
<li>This class extends <code>text_editor</code>. In this class, the <code>special_operation()</code> method is overridden to serve as an undo operation, allowing it to undo an infinite number of operations.</li>
</ul></li>
</ol>

<h3>Input</h3>

<pre tabindex="0" class="chroma"><code>[[&#39;special_text_editor&#39;], [&#39;write_line&#39;, &#39;special_operation&#39;, &#39;write_line&#39;], [&#39;World&#39;, &#39;Hello, &#39;]]
</code></pre>

<h3>Output</h3>

<code>&#34;Hello, World&#34;</code>

Create a text editor while displaying method overriding as a concept.

Text Editor With OOP

Design Patterns

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.

Data Preparation for Imbalanced Data

Let’s say we’re building a recommendation system for an e-commerce site, and we have two viable approaches: a collaborative filtering model that takes 2 days to implement, and a deep learning model that could boost accuracy by 10% but requires 3 weeks of engineering time.

How would you decide which solution to pursue, and what factors would you weigh in your decision?

Let’s say you’re working with a dataset of user-level events for a new product feature, and you notice issues like missing values, duplicate records, inconsistent data types, and some clear outliers. You need to prepare this data for training a machine learning model.

How would you design a data processing pipeline using Pandas and NumPy to clean and validate the data before modeling, and what steps would you take to ensure the pipeline is reusable and reliable for future use?

How would you build a reusable data cleaning pipeline for messy user event data?

User Event Data Pipeline

Data Pipelines

Imagine you run a pizza franchise, and you run into a problem with a lot of no-shows after customers place their order.

What features would you include in a model to try to predict a no-show?

What features would you include in a model to predict a no-show

Pizza No Show

Let’s say we’re launching a new version of the PayPal app, but due to a tight deadline, we don’t have time to run the entire suite of 300 automated test cases before release.

How would you handle this situation?

How would you approach limited testing time before a major app release?

Testing Constraints

A/B Testing

Given a string, write a function to determine if it is palindrome or not.

Note: A palindrome is a word/string that is read the same way forward as it is backward, e.g. <code>&#39;reviver&#39;</code>, <code>&#39;madam&#39;</code>, <code>&#39;deified&#39;</code> and <code>&#39;civic&#39;</code> are all palindromes, while <code>&#39;tree&#39;</code>, <code>&#39;music&#39;</code> and <code>&#39;person&#39;</code> are not palindromes.

Example:

Input:

<pre tabindex="0" class="chroma"><code>word1 = &#34;tree&#34;
word2 = &#34;radar&#34;
</code></pre>

Output:

<pre tabindex="0" class="chroma"><code>def is_palindrome(word1) -&gt; False
def is_palindrome(word2) -&gt; True
</code></pre>

Given a string, write a function to determine if it is palindrome or not.

String Palindromes

Let’s say we’ve built a neural network for image classification using NumPy, and we’re training it in a distributed environment with several workers. We notice that, even when using the same random seed, the loss curve looks erratic across different training runs, model weights diverge between workers, and training results aren’t consistent between single-node and distributed setups.

How would you go about diagnosing and resolving these distributed training inconsistencies to ensure stable and reproducible results?

How would you debug and fix inconsistencies in distributed NumPy neural network training?

Distributed Training Instability

Concurrent & Distributed Systems

When an interviewer asks a question along the lines of:

<ul>
<li>What would your current manager say about you? What constructive criticisms might he give?</li>
<li>What are your three biggest strengths and weaknesses you have identified in yourself?</li>
</ul>

How would you respond?

When asked about your strengths in an interview, what is an effective way to respond?

When asked about your strengths in an interview, what is an effective way to respond?

Your Strengths and Weaknesses I

Which of the following is an acceptable strategy when discussing weaknesses in an interview?

Which of the following is an acceptable strategy when discussing weaknesses in an interview?

Your Strengths and Weaknesses II

What do you tell an interviewer when they ask you what your strengths and weaknesses are?

Your Strengths and Weaknesses

Imagine you have been given an array of integers, and a query number <code>k</code>. Your task is to write a function that finds all the triplets in the array that sum up to the query number <code>k</code>.

Let’s illustrate this with some examples:

Example 1:

Input:

<pre tabindex="0" class="chroma"><code>nums = [1, 2, 3, 4, 5]
k = 6
</code></pre>

Output:

<pre tabindex="0" class="chroma"><code>count_triplets(nums, k) -&gt; 1
</code></pre>

In this case, there is only one triplet that sums up to 6: (1, 2, 3).

Example 2:

Input:

<pre tabindex="0" class="chroma"><code>nums = [10, 10, 20, 30, 40]
k = 60
</code></pre>

Output:

<pre tabindex="0" class="chroma"><code>count_triplets(nums, k) -&gt; 3
</code></pre>

In this case, there are three triplets that sum up to 60: (10, 10, 40), (10, 20, 30), and (10, 20, 30).
The value 10 can be retrieved as <code>nums[0]</code> or <code>nums[1]</code>, therefore resulting in two triplets with the same value.

Write a function that returns the number of triplets in the array that sum to `k`.

Triplet Counting

Let’s say we’re building a system for eBay to support multiple types of sellers: individuals, small businesses, and enterprise merchants. Each seller type shares core attributes like <code>seller_id</code> and <code>account_status</code>, but has unique features—enterprise merchants, for example, can manage up to 50 sub-accounts, while individuals have simpler profiles.

How would you decide between using composition or inheritance to model these seller types, and what trade-offs would you weigh in making your choice?

How would you decide between composition and inheritance when modeling different seller types on eBay?

Seller Type Modeling

Important: You must answer <a href="https://www.interviewquery.com/questions/array-operations-fixed-length-arrays-1">this question</a> before attempting this one. We recommend copying your previous solution and iterating the new solution with the added operations.

Create a more extensive <code>Array</code> class simulating the functionality of fixed-size arrays. The array’s fixed size is <code>6</code>. In this question, we will be implementing the deletion and search operations.

<ul>
<li><code>pop_back() -&gt; element</code> - removes the last element of the array. Returns the removed element. If the array is empty, return <code>None</code>.</li>

<li><code>pop_front() -&gt; element</code> - removes the first element of the array. Returns the removed element. If the array is empty, return <code>None</code>.</li>

<li><code>pop(index) -&gt; element</code> - removes the element at the specified index and returns it. When the index is not in range, raise an <code>IndexError</code>.</li>

<li><code>search(element) -&gt; index</code> - searches for the specified element and returns the index of the first instance if it exists inside the array. Returns <code>-1</code> if the element does not exist inside the array. Example: <code>Array([1, 2, 3]).search(2) -&gt; 1</code>.</li>

<li><code>__contains__(element) -&gt; bool</code> - checks if the element exists inside the array. Returns <code>True</code> if it exists and <code>False</code> if it does not.</li>

<li><code>abolish(element) -&gt; count</code> - removes all instances of an element and returns the count of all removed instances.</li>

<li><code>__eq__(other) -&gt; bool</code> - checks the equality of the array with another array called <code>other</code>. Raises a <code>TypeError</code> if <code>other</code> is not an instance of this custom <code>Array</code> class.</li>
</ul>

Note: All the basic addition operations must be implemented first. (Refer to Fixed Length Arrays 1)

You work as an ML engineer for a large company. The company wants to implement a machine learning system that utilizes facial recognition to facilitate employee clock-in, clock-out, and access to secure systems. The company also hires temporary contract consultants who need to be able to use the system.

How would you design this system?

Designing a secure and user-friendly facial recognition system for employee management while prioritizing privacy and ethical considerations

Distributed Authentication Model

Let’s say you’re trying to calculate churn for a subscription product.

You noticed that of all customers that bought subscriptions in January 2020, about 10% of them canceled their membership before their next cycle on February 1st.  

If you assume that your new customer acquisition is uniform throughout each month and that customer churn goes down by 20% month over month, what’s the expected churn rate in March for all customers that bought the product since January 1st?

What’s the expected churn rate in March for all customers that bought the product since January 1st?

Expected Churn

Product Sense & Metrics

Let’s say we’ve proposed a new analytics dashboard to a client, but after reviewing our design and data sources, the client says the solution doesn’t meet their needs and pushes back on several key features.

How would you approach the conversation to negotiate alternative solutions, and what steps would you take to ensure disagreements are handled constructively so we can still reach a successful outcome?

Let’s say we’re responsible for deploying a new version of Uber’s routing system, and the release includes both backend and frontend changes that may impact user workflows.

How would you design the deployment process to minimize risk, ensure clear communication with both clients and internal teams, and handle rollbacks if unexpected issues arise?

How would you structure a safe deployment process with effective stakeholder communication and rollback plans?

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

Pratt & Whitney Interview Questions

Pratt & Whitney Interview Guides

Pratt & Whitney Interview Questions

Challenge

Discussion & Interview Experiences

Discussion & Interview Experiences