Prepare for and practice interview questions from Imagine one technology & management, ltd..

Imagine one technology & management, ltd. Interview Questions

Imagine one technology & management, ltd. Interview Guides

Machine Learning

Why can the same machine learning algorithm produce different success rates on the same dataset

Same Algorithm Different Success

How would you build a model to detect if a post on a marketplace is talking about selling a gun?

Detecting Firearm Sales

Explain the key differences between classification models and regression models

Classification and Regression

Explain the difference between XGBoost and random forest and give an example where you would use one over the other

Xgboost vs Random Forest

Explain the relationship between PCA and K-means clustering

PCA and K-Means

Behavioral

Describing a data project and its challenges

Hurdles In Data Projects

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

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

Your Strengths and Weaknesses

Calculate the first touch attribution for each `user_id` that converted.

First Touch Attribution

Project Budget Error

Write a query to get the top five projects by budget-to-employee-count ratio

Employee Project Budgets

Count the number of users that like each user

Liker's Likers

Data Structures & Algorithms

Write a function str_map to determine if there exists a one-to-one correspondence between characters of string1 and string2

String Mapping

Write a function `sorting` from scratch to sort a list of strings in ascending alphabetical order

Sort Strings

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

Query Optimization

How would you diagnose and speed up a slow SQL query when system metrics look healthy?

Slow SQL Query

When an interviewer asks you a question along the lines of:

<ul>
<li>Why did you apply to our company?</li>
<li>What are you looking for in your next job?</li>
<li>What makes you a good fit for our company?</li>
</ul>

How should you respond?

When asked 'What are you looking for in your next job?' in an interview, how can you tie the company's employee benefits into your response?

When asked 'What are you looking for in your next job?' in an interview, how can you tie the company's employee benefits into your response?

Why Do You Want to Work With Us I

How can company values be used effectively in an interview when asked 'What makes you a good fit for our company?'

How can company values be used effectively in an interview when asked 'What makes you a good fit for our company?'

Why Do You Want to Work With Us II

When responding to the question 'Why did you apply to our company?' during an interview, what aspect should you highlight?

When responding to the question 'Why did you apply to our company?' during an interview, what aspect should you highlight?

Why Do You Want to Work With Us III

The schema below is for a retail online shopping company consisting of two tables, <code>attribution</code> and <code>user_sessions</code>.

<ul>
<li>The attribution table logs a session visit for each row.</li>

<li>If <code>conversion</code> is <code>true</code>, then the user converted to buying on that session.</li>

<li>The <code>channel</code> column represents which advertising platform the user was attributed to for that specific session.</li>

<li>Lastly the <code>user_sessions</code> table maps many to one session visits back to one user.</li>
</ul>

First touch attribution is defined as the channel with which the converted user was associated when they first discovered the website.

Calculate the first touch attribution for each <code>user_id</code> that converted. 

Example:

Input:

<code>attribution</code> table

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

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

<tr>
<td><code>channel</code></td>
<td>VARCHAR</td>
</tr>

<tr>
<td><code>conversion</code></td>
<td>BOOLEAN</td>
</tr>
</tbody>
</table>
<code>user_sessions</code> table

<table>
<thead>
<tr>
<th>column</th>
<th>type</th>
</tr>
</thead>

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

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

<tr>
<td><code>user_id</code></td>
<td>INTEGER</td>
</tr>
</tbody>
</table>
Example output:

<table>
<thead>
<tr>
<th><code>user_id</code></th>
<th><code>channel</code></th>
</tr>
</thead>

<tbody>
<tr>
<td>123</td>
<td>facebook</td>
</tr>

<tr>
<td>145</td>
<td>google</td>
</tr>

<tr>
<td>153</td>
<td>facebook</td>
</tr>

<tr>
<td>172</td>
<td>organic</td>
</tr>

<tr>
<td>173</td>
<td>email</td>
</tr>
</tbody>
</table>

There are two tables. One table is called <code>swipes</code> that holds a row for every Tinder swipe and contains a boolean column that determines if the swipe was a right or left swipe called <code>is_right_swipe</code>. The second is a table named <code>variants</code> that determines which user has which variant of an AB test.

Write a SQL query to output the average number of right swipes for two different variants of a feed ranking algorithm by comparing users that have swiped 10, 50, and 100 swipes in a <code>feed_change</code> experiment.

Note: Users have to have swiped at least 10 times to be included in the subset of users to analyze the mean number of right swipes.

Example:

Input:

<code>variants</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>experiment</code></td>
<td>VARCHAR</td>
</tr>

<tr>
<td><code>variant</code></td>
<td>VARCHAR</td>
</tr>

<tr>
<td><code>user_id</code></td>
<td>INTEGER</td>
</tr>
</tbody>
</table>
<code>swipes</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>user_id</code></td>
<td>INTEGER</td>
</tr>

<tr>
<td><code>swiped_user_id</code></td>
<td>INTEGER</td>
</tr>

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

<tr>
<td><code>is_right_swipe</code></td>
<td>BOOLEAN</td>
</tr>
</tbody>
</table>
Output:

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

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

<tr>
<td><code>mean_right_swipes</code></td>
<td>FLOAT</td>
</tr>

<tr>
<td><code>swipe_threshold</code></td>
<td>INTEGER</td>
</tr>

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

Write a SQL query to find the average number of right swipes for different ranking algorithms.

Swipe Precision

A/B Testing

We’re given two tables. One is named <code>projects</code> and the other maps employees to the projects they’re working on. 

We want to select the five most expensive projects by budget to employee count ratio. But let’s say that we’ve found a bug where there exist duplicate rows in the <code>employee_projects</code> table.

Write a query to account for the error and select the top five most expensive projects by budget to employee count ratio.

<h3>Schema</h3>

Input:

<code>projects</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>title</code></td>
<td>VARCHAR</td>
</tr>

<tr>
<td><code>state_date</code></td>
<td>DATETIME</td>
</tr>

<tr>
<td><code>end_date</code></td>
<td>DATETIME</td>
</tr>

<tr>
<td><code>budget</code></td>
<td>INTEGER</td>
</tr>
</tbody>
</table>
<code>employee_projects</code> table

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

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

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

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

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

<tr>
<td><code>budget_per_employee</code></td>
<td>FLOAT</td>
</tr>
</tbody>
</table>

<h3>Example</h3>

Input:

<code>projects</code> table

<table>
<thead>
<tr>
<th>id</th>
<th>title</th>
<th>start_date</th>
<th>end_date</th>
<th>budget</th>
</tr>
</thead>

<tbody>
<tr>
<td>1</td>
<td>party</td>
<td>2006-04-17 00:00:00</td>
<td>2006-05-19 00:00:00</td>
<td>25875</td>
</tr>

<tr>
<td>2</td>
<td>diversity</td>
<td>2017-08-09 00:00:00</td>
<td>2017-09-18 00:00:00</td>
<td>77867</td>
</tr>

<tr>
<td>3</td>
<td>integration</td>
<td>2005-06-29 00:00:00</td>
<td>2005-07-19 00:00:00</td>
<td>75987</td>
</tr>

<tr>
<td>4</td>
<td>testing</td>
<td>2009-01-12 00:00:00</td>
<td>2009-05-03 00:00:00</td>
<td>35946</td>
</tr>

<tr>
<td>5</td>
<td>launch</td>
<td>2005-05-02 00:00:00</td>
<td>2005-10-16 00:00:00</td>
<td>66292</td>
</tr>

<tr>
<td>6</td>
<td>meet</td>
<td>2001-09-09 00:00:00</td>
<td>2002-04-04 00:00:00</td>
<td>71243</td>
</tr>

<tr>
<td>7</td>
<td>payroll</td>
<td>2019-09-25 00:00:00</td>
<td>2020-05-05 00:00:00</td>
<td>97071</td>
</tr>

<tr>
<td>8</td>
<td>admin</td>
<td>2000-07-08 00:00:00</td>
<td>2000-09-10 00:00:00</td>
<td>24000</td>
</tr>

<tr>
<td>9</td>
<td>petronas</td>
<td>2016-07-01 00:00:00</td>
<td>2016-11-28 00:00:00</td>
<td>37088</td>
</tr>

<tr>
<td>10</td>
<td>bel</td>
<td>2005-08-03 00:00:00</td>
<td>2006-04-29 00:00:00</td>
<td>61937</td>
</tr>
</tbody>
</table>
<code>employee_projects</code> table

<table>
<thead>
<tr>
<th>project_id</th>
<th>employee_id</th>
</tr>
</thead>

<tbody>
<tr>
<td>1</td>
<td>7</td>
</tr>

<tr>
<td>1</td>
<td>9</td>
</tr>

<tr>
<td>2</td>
<td>6</td>
</tr>

<tr>
<td>3</td>
<td>1</td>
</tr>

<tr>
<td>3</td>
<td>2</td>
</tr>

<tr>
<td>4</td>
<td>5</td>
</tr>

<tr>
<td>5</td>
<td>3</td>
</tr>

<tr>
<td>5</td>
<td>4</td>
</tr>

<tr>
<td>6</td>
<td>8</td>
</tr>

<tr>
<td>8</td>
<td>10</td>
</tr>

<tr>
<td>9</td>
<td>11</td>
</tr>

<tr>
<td>10</td>
<td>12</td>
</tr>

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

<tr>
<td>9</td>
<td>11</td>
</tr>

<tr>
<td>6</td>
<td>8</td>
</tr>

<tr>
<td>4</td>
<td>5</td>
</tr>
</tbody>
</table>
Output:

<table>
<thead>
<tr>
<th>title</th>
<th>budget_per_employee</th>
</tr>
</thead>

<tbody>
<tr>
<td>diversity</td>
<td>77867</td>
</tr>

<tr>
<td>meet</td>
<td>71243</td>
</tr>

<tr>
<td>bel</td>
<td>61937</td>
</tr>

<tr>
<td>integration</td>
<td>37993.5</td>
</tr>

<tr>
<td>petronas</td>
<td>37088</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?

We’re given two tables. One is named <code>projects</code> and the other maps employees to the projects they’re working on.

Write a query to get the top five most expensive projects by budget to employee count ratio.

Note: Exclude projects with 0 employees. Assume each employee works on only one project.

<code>projects</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>title</code></td>
<td>VARCHAR</td>
</tr>

<tr>
<td><code>state_date</code></td>
<td>DATETIME</td>
</tr>

<tr>
<td><code>end_date</code></td>
<td>DATETIME</td>
</tr>

<tr>
<td><code>budget</code></td>
<td>INTEGER</td>
</tr>
</tbody>
</table>
<code>employee_projects</code> table

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

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

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

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

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

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

How would you explain what a p-value is to someone who is not technical?

What does a p-value in a statistical test represent?

What does a p-value in a statistical test represent?

P-value to a Layman I

In a statistical test, how does a low p-value (less than 0.05) influence our decision about the null hypothesis?

In a statistical test, how does a low p-value (less than 0.05) influence our decision about the null hypothesis?

P-value to a Layman II

Explain what a p-value is to someone who is not technical

P-value to a Layman

Statistics

Let’s say you are designing a marketplace for your website.

Selling firearms is prohibited by your website’s Terms of Service Agreement (not to mention the laws of your country). To this end, you want to create a system that can automatically detect if a listing on the marketplace is selling a gun.

How would you go about doing this?

While designing a system to detect firearms listings on a marketplace website, what metrics could be used to minimize both false positives and false negatives?

While designing a system to detect firearms listings on a marketplace website, what metrics could be used to minimize both false positives and false negatives?

Given two strings, <code>string1</code> and <code>string2</code>, write a function <code>str_map</code> to determine if there exists a one-to-one correspondence (bijection) between the characters of <code>string1</code> and <code>string2</code>.

For the two strings, our correspondence must be between characters in the same position/index.

Example 1:

Input:

<pre tabindex="0" class="chroma"><code>string1 = &#39;qwe&#39;
string2 = &#39;asd&#39;

string_map(string1, string2) == True

# q = a, w = s, and e = d
</code></pre>

Example 2:

Input:

<pre tabindex="0" class="chroma"><code>string1 = &#39;donut&#39;
string2 = &#39;fatty&#39;

string_map(string1, string2) == False
# cannot map two distinct characters to two equal characters
</code></pre>

Example 3:

Input:

<pre tabindex="0" class="chroma"><code>string1 = &#39;enemy&#39;
string2 = &#39;enemy&#39;

string_map(string1, string2) == True
# there exists a one-to-one correspondence between equivalent strings
</code></pre>

Example 4:

Input:

<pre tabindex="0" class="chroma"><code>string1 = &#39;enemy&#39;
string2 = &#39;ymene&#39;

string_map(string1, string2) == False
# since our correspondence must be between characters of the same index, this case returns &#39;False&#39; as we must map e = y AND e = e
</code></pre>

Given a list of strings, write a function, <code>sorting</code> from scratch to sort the list in ascending alphabetical order.

Notes:
- Do NOT use the built-in <code>sorted</code> function
- Return the new sorted list, rather than modifying the list in-place

Bonus: Have your solution be \(O(n log(n))\).

Example:

Input:

<pre tabindex="0" class="chroma"><code>array = [&#34;apple&#34;, &#34;cat&#34;, &#34;banana&#34;, &#34;zoo&#34;, &#34;football&#34;]
</code></pre>

Output:

<pre tabindex="0" class="chroma"><code>def sorting(array) -&gt; [&#34;apple&#34;, &#34;banana&#34;, &#34;cat&#34;, &#34;football&#34;, &#34;zoo&#34;]
</code></pre>

Why would the same machine learning algorithm generate different success rates using the same dataset?

Note: When they ask us an ambiguous question, we need to gather context and restate it in a way that’s clear for us to answer.

Let’s say that you work on the revenue forecasting team at a company like Facebook.

An executive comes to you asking about how much revenue Facebook will make in the coming year.

How would you forecast revenue for the next year?

How would you forecast revenue for the next year

Forecasting New Year Revenue

Forecasting & Time Series

As a data scientist at a real estate analytics startup, you’re tasked with building a predictive model for home prices in a large metropolitan area. During your exploratory data analysis, you notice that the <code>home_prices</code> column in your dataset is heavily right-skewed (i.e., there are many lower-priced homes and a few extremely high-priced ones).

How might this skewness impact your modeling approach, and what steps would you consider to address it before training your model?

Bonus: If, instead, you discover that the target variable (<code>home_prices</code>) is heavily left-skewed (i.e., there are many high-priced homes and a few low-priced outliers), what would you do differently, if anything?

You're building a model to predict home prices and you see that the prices in your training data are right-skewed. How would you handle this situation?

Skewed Pricing

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

Build a \(k\) Nearest Neighbors classification model from scratch with the following conditions:

<ul>
<li>Use Euclidian distance (aka, the “2 norm”) as your closeness metric</li>
<li>Your function should be able to handle data frames of arbitrary many rows and columns</li>
<li>If there is a tie in the class of the \(k\) nearest neighbors, rerun the search using \(k-1\) neighbors instead</li>
<li>You may use <code>pandas</code> and <code>numpy</code> but NOT <code>scikit-learn</code></li>
</ul>

Example:

Input:

<pre tabindex="0" class="chroma"><code>k = 5
new_point = [0.5,-2,8]
print(data)
...
 Var1 Var2 Var3 Target
0 -3.279536 3.362223 2.847892 2
1 -0.791565 1.742475 2.151587 2
2 -0.785992 -0.938681 -0.459770 0
3 -1.068190 1.461051 0.127130 3
4 -0.367568 -0.870240 -0.225734 0
.. ... ... ... ...
95 -1.327175 1.971085 -0.690689 2
96 -3.203714 1.847649 0.778901 2
97 -0.587640 0.647458 2.094385 2
98 0.363644 -0.509795 2.514191 1
99 -0.673498 2.955285 2.102122 4

[100 rows x 4 columns]
</code></pre>

Output:

<pre tabindex="0" class="chroma"><code>def kNN(k, new_point, data) -&gt; 2
</code></pre>

Build a k Nearest Neighbors classification model from scratch.

KNN From Scratch

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 data engineer at Fidelity Investments, and you’re running a SQL query on a cloud-based data warehouse. All cluster resources and network health metrics look normal, but the query is still taking over 10 minutes to complete.

How would you go about diagnosing and improving the performance of this query?

A dating website’s schema is represented by a table of people that like other people. The table has three columns. One column is the <code>user_id</code>, another column is the <code>liker_id</code> which is the <code>user_id</code> of the user doing the liking, and the last column is the <code>date time</code> that the like occurred.

Write a query to count the number of liker’s likers (the users that like the likers) if the liker has one.

Example:

Input:

<code>likes</code> table

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

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

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

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

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

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

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

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

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

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?

Explain the difference between the XGBoost and random forest algorithms and give an example where you would use one over the other.

Why might a data scientist choose to use XGBoost instead of Random Forest for a particular machine learning task?

Why might a data scientist choose to use XGBoost instead of Random Forest for a particular machine learning task?

XGBoost vs Random Forest: Choice

What’s the relationship between PCA and K-means clustering?

What does the variable “k” in k-means clustering refer to?

What does the variable "k" in k-means clustering refer to?

Input of K-means

What does the backpropagation algorithm do in the context of neural networks? What is the informal intuition behind the algorithm? What are some drawbacks of the algorithm compared to other optimization methods?

Bonus: Formally derive the backpropagation algorithm and prove that it does what it claims to do.

Explain what backpropagation does, its informal intuition, drawbacks, and derive/prove it

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

Imagine one technology & management, ltd. Interview Questions

Imagine one technology & management, ltd. Interview Guides

Imagine one technology & management, ltd. Interview Questions

Challenge

Discussion & Interview Experiences

Discussion & Interview Experiences