Prepare for and practice interview questions from Torc Robotics.

Torc Robotics Interview Questions

Torc Robotics Interview Guides

Data Structures & Algorithms

Write a function n_frequent_words that returns the top N frequent words and their frequencies, and state its run-time

Get Top N Frequent Words

Given an array and a target integer, write a function that returns the indices of two integers in the array that add up to the target integer.

Target Indices

Write a function digit_accumulator that returns the sum of every digit in a string

Digit Accumulator

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

The Brackets Problem

Write a function to return the value of the nearest node that is a parent to both nodes.

Nearest Common Ancestor

Machine Learning

Lasso vs Ridge

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

Bagging vs Boosting

What are some machine learning methods for building this system

FAQ Matching

How to model merchant acquisition in a new market?

Merchant Acquisition

Explain the key differences between classification models and regression models

Classification and Regression

Select the 2nd highest salary in the engineering department

2nd Highest Salary

Duplicate Rows

Write a query to get the duration in minutes of each ride longer than two hours, sorted descending

Time Difference

Write a query to find how much time each plane spent in the air each day in minutes rounded down

Total Time in Flight

Write a query that outputs a random manufacturer's name with an equal probability of selecting any name.

Uniform Car Maker

Product Sense & Metrics

What would you investigate to diagnose what's happening?

WAU vs Open Rates

Google Maps Improvement

Unified Inbox

What are the top five metrics you would start tracking to understand the health of Google Docs

Docs Metrics

The role of A/B testing in measuring the success rate of an analytics experiment

Success Measurement

Behavioral

Describing a data project and its challenges

Hurdles In Data Projects

Prioritized debt reduction, process improvement, and a focus on maintainability for fintech efficiency

Decreasing Tech Debt

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

Why Do You Want to Work With Us

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

Client Solution Pushback

Making data-driven insights actionable for those without technical expertise

Simple Explanations

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>

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?

Say you are tasked with designing a data mart or data warehouse for a new online retailer.

How would you design the system?

Note: Sketch a star schema to explain your design.

Design a data warehouse for a new online retailer

Retailer Data Warehouse

Data Modeling

Let’s say that you’re a data scientist on the engagement team. A product manager comes up to you and says that the weekly active users metric is up 5% but email notification open rates are down 2%.

What would you investigate to diagnose what’s happening?

Email Open Rate

Let’s say you work in air traffic control. You are given the table below containing information on flights between two cities.

Write a query to find out how much time, in minutes (rounded down), each plane spent in the air each day.

Note: Both cities are in the same time zone, so you do not need to worry about converting time zones.

<code>flights</code> table

Example:

Input:

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

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

<tr>
<td><code>destination_location</code></td>
<td>VARCHAR</td>
</tr>

<tr>
<td><code>source_location</code></td>
<td>VARCHAR</td>
</tr>

<tr>
<td><code>plane_id</code></td>
<td>INTEGER</td>
</tr>

<tr>
<td><code>flight_start</code></td>
<td>DATETIME</td>
</tr>

<tr>
<td><code>flight_end</code></td>
<td>DATETIME</td>
</tr>
</tbody>
</table>
Output:

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

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

<tr>
<td><code>calendar_day</code></td>
<td>VARCHAR</td>
</tr>

<tr>
<td><code>time_in_min</code></td>
<td>INTEGER</td>
</tr>
</tbody>
</table>
Please give <code>calendar_day</code> in the format <code>YYYY-MM-DD</code>

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

<ol>
<li>Given an example paragraph string and an integer <code>N</code>, write a function <code>n_frequent_words</code> that returns the top <code>N</code> frequent words in the posting and the frequencies for each word.</li>

<li>What’s the function run-time?</li>
</ol>

Example:

Input:

<pre tabindex="0" class="chroma"><code>posting = &#34;&#34;&#34;
Herbal sauna uses the healing properties of herbs in combination with distilled water. 
The water evaporates and distributes the effect of the herbs throughout the room. 
A visit to the herbal sauna can cause real miracles, especially for colds. 
&#34;&#34;&#34; 
n = 3
</code></pre>

Output:

<pre tabindex="0" class="chroma"><code>n_frequent_words(posting,N) = [
 (&#39;the&#39;, 6), 
 (&#39;herbal&#39;, 2), 
 (&#39;sauna&#39;, 2),
]
</code></pre>

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/>

You work as a data scientist at Cars.com. The company is launching a new personalized car recommendation engine that must support both real-time and batch inference, adapt to changing user preferences, and enable frequent model updates via CI/CD pipelines.

How would you design the end-to-end architecture for this system, ensuring it can robustly detect and respond to data distribution shifts while maintaining seamless CI/CD for machine learning models?

Design an end-to-end ML system for personalized car recommendations at Cars.com, supporting real-time and batch inference, data distribution shift detection, and CI/CD for frequent model updates.

Car Recommendation Architecture

Let’s say that you’re the PM on Google Maps.

<ol>
<li>How would you improve Google Maps?</li>

<li>What metrics would you check to see if your feature improvements are successful?</li>
</ol>

You work at a ride-sharing company and are given a table containing information about the rides of users.

Write a query to get the duration (in minutes) of each ride longer than two hours. Sort the results by duration in descending order.

<code>rides</code> table:

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

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

<tr>
<td><code>passenger_user_id</code></td>
<td>INTEGER</td>
</tr>

<tr>
<td><code>start_dt</code></td>
<td>DATETIME</td>
</tr>

<tr>
<td><code>end_dt</code></td>
<td>DATETIME</td>
</tr>

<tr>
<td><code>distance</code></td>
<td>FLOAT</td>
</tr>

<tr>
<td><code>ride_region</code></td>
<td>VARCHAR</td>
</tr>

<tr>
<td><code>is_completed</code>	</td>
<td>INTEGER</td>
</tr>
</tbody>
</table>
Output:

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

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

You are given a binary tree of unique positive numbers. Each node in the binary tree is represented as a dictionary with the following keys:

<ul>
<li><code>data</code>: integer value stored in the node</li>
<li><code>left</code>: reference to the left child (or <code>None</code>)</li>
<li><code>right</code>: reference to the right child (or <code>None</code>)</li>
</ul>

<pre tabindex="0" class="chroma"><code>node = {
 &#34;data&#34;: 6,
 &#34;left&#34;: {
 &#34;data&#34;: 3,
 &#34;left&#34;: {...},
 &#34;right&#34;: {...}
 },
 &#34;right&#34;: {
 &#34;data&#34;: 9,
 &#34;left&#34;: {...},
 &#34;right&#34;: {...}
 }
}
</code></pre>

Given two node values as input (<code>value1</code> and <code>value2</code>), write a function to return the value of the nearest common ancestor (lowest node in the tree that has both nodes as descendants).

Note: If one of the nodes doesn’t exist in the tree, return <code>-1</code>.

Example:

Input:

<pre tabindex="0" class="chroma"><code># Diagram of the binary tree
&#39;&#39;&#39;
 6 
 / \ 
 3 9 
 / \
 2 11
 / \
 5 8
&#39;&#39;&#39;
value1 = 8
value2 = 2
</code></pre>

Output:

<pre tabindex="0" class="chroma"><code>common_ancestor(root,value1,value2) -&gt; 3
</code></pre>

Explanation:

<ul>
<li>Ancestors of <code>8</code>: <code>11 → 3 → 6</code></li>
<li>Ancestors of <code>2</code>: <code>3 → 6</code></li>
<li>Common ancestors: <code>3</code> and <code>6</code></li>
<li>The nearest common ancestor is <code>3</code>.</li>
</ul>

Say you work for an e-commerce website that receives clickstream data from Kafka on a daily basis. You’re asked to design a data analytics solution for storing and querying raw data, with a two-year retention period.

The current data volume is around 600 million events daily.

What kind of data analytics solution would you design, keeping costs in mind?

Design a cost-conscious analytics solution to store and query daily Kafka clickstream data with two-year retention

Clickstream Data

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…

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

The support team at your company is trying to analyze ticketing data to improve response times. They have noticed that a significant number of tickets have no assigned agent, making it difficult to track the progress of these tickets.

Write a SQL query to count the total number of tickets, the number of tickets assigned to agents, and the number of tickets that are not assigned to any agent.

Note: Tickets that are not assigned to any agent have <code>NULL</code> in their <code>agent_id</code> field.

Example:

Input:

<code>tickets</code> table

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

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

<tr>
<td>issue</td>
<td>VARCHAR</td>
</tr>

<tr>
<td>agent_id</td>
<td>INTEGER</td>
</tr>

<tr>
<td>created_at</td>
<td>TIMESTAMP</td>
</tr>

<tr>
<td>updated_at</td>
<td>TIMESTAMP</td>
</tr>
</tbody>
</table>
Output:

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

<tbody>
<tr>
<td>total_tickets</td>
<td>INTEGER</td>
</tr>

<tr>
<td>tickets_with_agent</td>
<td>INTEGER</td>
</tr>

<tr>
<td>tickets_without_agent</td>
<td>INTEGER</td>
</tr>
</tbody>
</table>

Count total tickets, tickets with agent assignment, and tickets without agent assignment.

Ticket Agent Analysis

Suppose there exists a new airline named Jetco that flies domestically across North America. Jetco recently had a study commissioned that tested the boarding time of every airline, and it came out that Jetco had the fastest average boarding times of any airline.

What factors could have biased this result, and what would you look into?

What could potentially bias the average boarding time in favor of an airline that only flies domestically?

What could potentially bias the average boarding time in favor of an airline that only flies domestically?

Boarding Times Bias

A new airline came out as the fastest average boarding times compared to other airlines.  What factors could have biased this result and what would you look into?

Statistics

You’re a data scientist at YouTube. Leadership wants to revisit how the platform measures success, not for a specific new feature, but for the product as a whole given its current state.

How would you define and measure success for YouTube? Identify your primary and secondary metrics, explain your rationale for each, and discuss how you would choose the right thresholds for any engagement-based metrics you propose.

Given an array and a target integer, write a function <code>sum_pair_indices</code> that returns the indices of two integers in the array that add up to the target integer. If not found, just return an empty list.

Note: Can you do it on \(O(n)\) time?

Note: Even though there could be many solutions, only one needs to be returned.

Example 1:

Input:

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

Output:

<pre tabindex="0" class="chroma"><code>def sum_pair_indices(array, target) -&gt; [0 3] or [1 2]
</code></pre>

Example 2:

Input:

<pre tabindex="0" class="chroma"><code>array = [3]
target = 6 
</code></pre>

Output:

<pre tabindex="0" class="chroma"><code>Do NOT return [0 0] as you can&#39;t use an index twice.
</code></pre>

Given a table of <code>cars</code> with columns <code>id</code> and <code>make</code>, write a query that outputs a random manufacturer’s name with an equal probability of selecting any name.

Input:

<code>cars</code> table

<table>
<thead>
<tr>
<th><code>id</code></th>
<th><code>make</code></th>
</tr>
</thead>

<tbody>
<tr>
<td>1</td>
<td>Ford</td>
</tr>

<tr>
<td>2</td>
<td>Toyota</td>
</tr>

<tr>
<td>3</td>
<td>Toyota</td>
</tr>

<tr>
<td>4</td>
<td>Honda</td>
</tr>

<tr>
<td>5</td>
<td>Honda</td>
</tr>

<tr>
<td>6</td>
<td>Honda</td>
</tr>
</tbody>
</table>
Output:

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

<tbody>
<tr>
<td><code>make</code></td>
<td>Text</td>
</tr>
</tbody>
</table>

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?

Given a <code>users</code> table, write a query to return only its duplicate rows.

Example:

Input:

<code>users</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>

<tr>
<td><code>created_at</code></td>
<td>DATETIME</td>
</tr>
</tbody>
</table>

You are given a <code>string</code> that represents some floating-point number. Write a function, <code>digit_accumulator</code>, that returns the sum of every digit in the <code>string</code>.

Example:

Input:

<pre tabindex="0" class="chroma"><code>s = &#34;123.0045&#34;
</code></pre>

Output:

<pre tabindex="0" class="chroma"><code>def digit_accumulator(s) -&gt; 15

Since 
1 + 2 + 3 + 0 + 0 + 4 + 5 = 15
</code></pre>

Let’s say that you’re in charge of Square’s small business division. 

The CEO wants to hire a customer success manager to help manage a new software product. Another executive thinks that it’s worth just instituting a free trial instead.

What would be your recommendation on utilizing a customer success manager versus just a free trial to get new or existing customers to use the new product?

What would be your recommendation on utilizing a customer success manager versus just a free trial to get new or existing customers to use the new product

Customer Success vs. Free Trial

Business Case

You’re a data analyst at a global CPG distributor that tracks daily stock availability across ~80 countries for the brands in its portfolio. The operations team wants to triage which brands are exposed to systemic supply-chain risk, which have only localized issues, and which are running clean so they know where to spend investigative cycles.

You have access to a daily inventory table:

<code>inventory_status</code>

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

<tbody>
<tr>
<td><code>date</code></td>
<td>DATE</td>
<td>The day the snapshot was taken</td>
</tr>

<tr>
<td><code>country</code></td>
<td>VARCHAR</td>
<td>ISO country code where stock is held</td>
</tr>

<tr>
<td><code>company</code></td>
<td>VARCHAR</td>
<td>Brand in the portfolio</td>
</tr>

<tr>
<td><code>product</code></td>
<td>VARCHAR</td>
<td>SKU identifier</td>
</tr>

<tr>
<td><code>is_out_of_stock</code></td>
<td>BOOLEAN</td>
<td><code>TRUE</code> if the SKU was out of stock at any point that day in that country</td>
</tr>
</tbody>
</table>
Over the trailing 30 days, every company in the portfolio falls into one of three buckets:

<ol>
<li>Resilient — no stock-outs anywhere.</li>
<li>Localized risk — stock-outs occurred, but only ever within a single country.</li>
<li>Systemic risk — stock-outs occurred across multiple countries.</li>
</ol>

Given this bucket distribution, propose three business hypotheses the operations team should investigate to act on these results. For each hypothesis:

<ul>
<li>State the mechanism you think is driving the pattern.</li>
<li>Describe the follow-up analysis that would confirm or kill it.</li>
</ul>

Classify companies as resilient, localized, or systemic stock-out risk, then propose three falsifiable hypotheses explaining the distribution

Out of Stock Inventory

Let’s say that we want to build a chatbot system for frequently asked questions. Whenever a user writes a question, we want to return the closest answer from a list of FAQs.

What are some machine learning methods for building this system?

Let’s say we’re designing a deep learning pipeline to process millions of customer support chat logs, and we want to use a Transformer encoder layer built in PyTorch to extract useful representations for downstream tasks like sentiment classification.

How would you approach building this Transformer encoder layer from scratch—what are the essential components, how would you implement them in PyTorch, and why are elements like residual connections and layer normalization critical for stable and effective training?

How would you build and justify the components of a Transformer encoder layer in PyTorch for large-scale text data?

Transformer Encoder Layer

AI & Agentic Systems

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

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

A robot has been designed to navigate a two-dimensional 4x4 matrix by only moving forward or turning right when blocked by a wall of the matrix. Its starting position is in the top left corner of the matrix, denoted by (0,0), and the robot’s final destination is the bottom right corner.

Determine the full path of the robot before it hits the final destination or starts repeating the path.

The ‘walls’ of the matrix are either one of the four borders of the map or any block found within.

A ‘Block’ is donated by having 1 in the cell corresponding to it where an empty cell is having ‘0’.

Example 1:

<pre tabindex="0" class="chroma"><code>#Input
robot_map = [
 [0,0,0,0],[0,0,1,0],[0,1,0,0],[0,0,0,0]]

#Output
robot_path(robot_map) -&gt; [(0,0),(0,1),(0,2),(0,3),(1,3),(2,3),(3,3)]
</code></pre>

As seen in these diagrams, the robot’s path starts with this sequence of steps:

<ol>
<li>Goes forward three times and hits the wall. Turns right.</li>
<li>Goes forward three times (orientation has changed), 3. Reaches the final destination, and stops.</li>
</ol>

<img src="/s3/questions/a251a8da-2906-40de-9fe8-7f15f367fa34.png" alt="image"/>

Example 2:

<pre tabindex="0" class="chroma"><code>#Input
robot_map = [
 [0,0,1,0],[0,0,0,0],[0,1,0,0],[0,0,0,0]]

#Output
robot_path(robot_map) -&gt; [(0,0),(0,1),(1,1),(1,0)]

</code></pre>

Here, the robot moves forward one space, but hits the wall. After turning right, another wall is hit, starting a loop.

<img src="/s3/questions/34e81379-4514-472c-84a0-f31a4721f5b5.png" alt="image"/>

Determine the full path of the robot before it hits the final destination or starts repeating the path.

Walking Robot

Find if there is a path from a starting point to an ending point in a walled maze

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

Torc Robotics Interview Questions

Torc Robotics Interview Guides

Torc Robotics Interview Questions

Challenge

Discussion & Interview Experiences

Discussion & Interview Experiences