40 Analytics Interview Questions

Are you prepared for questions like 'Have you ever used machine learning algorithms? If so, which ones and for what purposes?' and similar? We've collected 40 interview questions for you to prepare for your next Analytics interview.

Have you ever used machine learning algorithms? If so, which ones and for what purposes?

Yeah, I've used a bunch of machine learning algorithms for different projects. For instance, I've used linear regression and decision trees for predictive analytics, such as forecasting sales and predicting customer churn. I've also implemented clustering algorithms like K-means for market segmentation to better understand customer groups. Additionally, neural networks have come in handy for image recognition tasks in a couple of projects. Each algorithm really has its own strengths depending on the problem you're trying to solve.

How do you stay updated with the latest trends and tools in data analytics?

I like to immerse myself in a mix of online courses, blogs, webinars, and industry news sites. Websites like Coursera or Udacity offer advanced courses that keep my skills sharp and up-to-date. Additionally, I follow key influencers on LinkedIn and regularly check out articles from platforms like Medium and Towards Data Science. Attending conferences and meetups, even virtually, is another great way to stay in touch with the community and latest innovations.

Can you explain the difference between supervised and unsupervised learning?

Supervised learning involves training a model on a labeled dataset, meaning each training example is paired with an output label. You use these labels to teach the model and then evaluate its predictions on new, unseen data. It's typically used for tasks like classification (categorizing emails as spam or not spam) and regression (predicting house prices).

Unsupervised learning, on the other hand, deals with unlabeled data. The model tries to learn the patterns and structure from the data without any explicit instructions on what to predict. Common applications include clustering (grouping customers with similar purchasing habits) and dimensionality reduction (simplifying data for easier visualization).

How do you handle missing data in a dataset?

Handling missing data can be tackled in several ways depending on the context and the dataset. One common method is to simply remove the rows or columns that contain missing values, especially if they represent a small fraction of the total data. This approach can be quick but may lead to loss of valuable information.

Another approach is to impute missing values. You can use statistical methods such as mean, median, or mode imputation, where you replace missing values with the average, median, or most frequent value of that column. For more complex scenarios, advanced methods like regression imputation, K-nearest neighbors, or even machine learning algorithms can be used to predict and fill in the missing values based on the rest of the data.

Sometimes, missing values themselves can carry information. In such cases, you might encode missingness as a separate category, especially for categorical variables. The choice of technique depends on the data and how you intend to use it, but the key is to always know the implications of your chosen method on the analysis outcome.

What is your experience with SQL? Can you write a query to find the top 5 products by sales?

I've worked with SQL extensively in past roles, primarily for extracting and manipulating data from relational databases. Crafting queries to generate reports and insights is a regular part of my workflow. For finding the top 5 products by sales, you might write something like this:

sql SELECT product_name, SUM(sales) as total_sales FROM sales_table GROUP BY product_name ORDER BY total_sales DESC LIMIT 5;

This query groups the sales data by product, sums up the sales for each product, orders them by total sales in descending order, and then limits the result to the top 5.

What's the best way to prepare for a Analytics interview?

Seeking out a mentor or other expert in your field is a great way to prepare for a Analytics interview. They can provide you with valuable insights and advice on how to best present yourself during the interview. Additionally, practicing your responses to common interview questions can help you feel more confident and prepared on the day of the interview.

What is the purpose of A/B testing and how is it implemented?

A/B testing is used to compare two versions of something to determine which one performs better. The goal is to make data-driven decisions by experimenting with variables like website layouts, ad campaigns, or product features. By showing two different groups of users different versions, you can analyze which version leads to better engagement or conversion rates.

Implementation typically involves splitting the audience into two groups randomly. One group sees version A (the control) while the other group sees version B (the variation). You'll then collect performance data on metrics that matter to you, like click-through rates or conversion rates. After gathering enough data to reach statistical significance, you'll analyze the results to see which version performed better, allowing you to make informed decisions moving forward.

Describe Principal Component Analysis and its applications.

Principal Component Analysis (PCA) is a dimensionality reduction technique often used to transform a large set of variables into a smaller one that still contains most of the information. PCA works by identifying the directions, called principal components, along which the variation in the data is maximum. The first principal component is the direction of the greatest variance, the second is orthogonal to the first and accounts for the next greatest variance, and so on.

One common application of PCA is in image compression. By transforming the image data into principal components, you can store only the most significant components and thus reduce the amount of data to be stored, while still maintaining the essential features of the image. Another application is in exploratory data analysis; PCA can help visualize the structure of high-dimensional data by reducing it to two or three dimensions. It's also widely used for noise reduction, where it can help filter out the "noise" from the data, improving the performance of machine learning algorithms.

How would you approach building a predictive model?

When building a predictive model, I typically start with a clear definition of the problem and the objective. Understanding what I'm trying to predict and how the predictions will be used is crucial. Next, I gather and prepare the data, which often involves cleaning, handling missing values, and transforming variables to make sure the dataset is suitable for modeling.

I then explore different algorithms and evaluate their performance through cross-validation and other metrics. Often, I split the data into training and testing subsets to ensure the model generalizes well. Once I select the best model, I fine-tune it using techniques like hyperparameter optimization. Finally, I validate the model with real-world data or through other robust testing methods and monitor its performance over time to ensure it continues to work well.

Explain the difference between correlation and causation.

Correlation means that there is a relationship or pattern between the values of two variables. In other words, when one variable changes, the other tends to change in a specific way. However, this doesn't necessarily mean that one variable is causing the change in the other. Causation, on the other hand, implies that one variable directly affects or causes changes in another.

For instance, ice cream sales and drowning incidents might both increase during the summer, showing a correlation. But this doesn't mean that buying ice cream causes drowning incidents—both are likely influenced by the hotter weather, which is a third factor at play. Identifying causation usually requires deeper investigation, such as controlled experiments or longitudinal studies, to rule out other variables and establish a direct link.

Explain the concept of overfitting and how you can avoid it.

Overfitting occurs when a model learns the noise and details in the training data to the extent that it negatively impacts its performance on new data. Essentially, the model becomes too complex and tailored to the specific data it's been trained on, and as a result, it fails to generalize to unseen data.

To avoid overfitting, you can use techniques such as cross-validation, which involves dividing your data into subsets to ensure that the model performs well on different samples of the data. Additionally, you can employ regularization methods like L1 or L2 regularization to penalize overly complex models. Simplifying the model by reducing the number of features through feature selection or dimensionality reduction techniques like PCA can also help mitigate overfitting. Lastly, increasing the amount of training data, if possible, tends to improve the model's ability to generalize.

Describe a time when you used data analysis to solve a business problem.

In my previous role, our sales department was seeing a decline, and they couldn't figure out why. So, I dove into the sales data from the past year, segmenting it by product lines, regions, and customer demographics. I found that a particular product line was underperforming in a specific region due to an underestimated local competitor. Armed with this insight, we adjusted our marketing strategy to highlight key differentiators and ran targeted promotions. Within three months, sales in that region rebounded significantly, saving both the product line and boosting our overall numbers.

How do you ensure the quality and accuracy of your data?

I start by performing basic data cleaning, which includes removing duplicates, handling missing values, and correcting any inconsistencies. Validation checks like range and format validation are also crucial. I also often cross-verify data with source systems to ensure it matches up. Using tools like SQL and Python for automated checks can make this process more efficient.

Can you explain what a p-value is and how it is used in hypothesis testing?

A p-value is a measure that helps you determine the significance of your results in hypothesis testing. It essentially helps you figure out the likelihood that your observed data would occur if the null hypothesis were true. In simpler terms, it tells you whether your findings are out of the ordinary or just part of random variation.

If the p-value is very small (usually less than 0.05), it suggests that the observed data is unlikely under the null hypothesis, leading you to reject the null hypothesis in favor of the alternative hypothesis. On the other hand, a larger p-value indicates that the observed data is consistent with the null hypothesis, and you don't have enough evidence to reject it. It's a useful way to quantify the strength of your evidence against the null hypothesis.

What tools and software are you proficient in for data analysis?

I've had extensive experience with several tools and software for data analysis. I'm very proficient with Python, using libraries like Pandas, NumPy, and Matplotlib for data manipulation and visualization. Additionally, I've worked a lot with SQL for database querying and have used R for statistical analysis, especially in academic and research settings.

On the business intelligence side, I've used tools like Tableau and Power BI to create interactive dashboards and reports. For more advanced data manipulation and machine learning, I've leveraged platforms like Apache Spark and TensorFlow.

Describe a situation where you had to communicate complex data insights to a non-technical stakeholder.

I once had to present key findings from a customer segmentation analysis to our marketing team, who were not very tech-savvy. Instead of diving into the technical details, I focused on telling a story with the data. I used clear visualizations like bar graphs and pie charts, which are easier to understand at a glance, and related the data back to tangible business outcomes, such as which customer segments were most likely to respond to particular marketing strategies. This approach helped bridge the gap between the data analysis and the actionable insights the team needed.

Have you used Python or R for data analysis? Describe the libraries you commonly use.

Yes, I've used both Python and R for data analysis. In Python, I frequently use libraries like Pandas for data manipulation, NumPy for numerical operations, and Matplotlib or Seaborn for data visualization. For more advanced analytics, I often turn to Scikit-learn for machine learning tasks.

In R, I typically rely on the Tidyverse collection of packages, like dplyr for data manipulation and ggplot2 for visualization. For statistical modeling, I often use packages like caret or glmnet. Both languages have their strengths and are pretty versatile in handling a wide range of data analysis tasks.

Can you explain the cross-validation technique?

Cross-validation is a strategy to assess the generalizability and robustness of a statistical model. One common approach is k-fold cross-validation, where the data is divided into k equally sized folds. The model is trained on k-1 of those folds and validated on the remaining one. This process is repeated k times, each time with a different fold as the validation set. The final performance metric is then averaged across all k iterations, providing a more reliable estimate of the model's performance compared to using a single train-test split. This helps to ensure that the model isn’t just overfitting to a specific subset of the data.

What are your favorite data visualization tools and why?

I really enjoy using Tableau and Python's Seaborn library. Tableau is fantastic because it allows for quick, interactive dashboards that are user-friendly and can communicate insights effectively to non-technical stakeholders. On the other hand, Seaborn, built on top of Matplotlib, offers a more programmable approach, giving you the flexibility to create highly customized visualizations, which is great for in-depth analysis and reporting.

How would you handle an imbalanced dataset?

One effective way to handle an imbalanced dataset is to use techniques like resampling. This includes undersampling the majority class or oversampling the minority class to balance the dataset. Methods like SMOTE (Synthetic Minority Over-sampling Technique) can be particularly useful because they create synthetic samples rather than just duplicating existing ones.

Additionally, I might use performance metrics better suited for imbalanced datasets, like precision, recall, or the F1-score, rather than accuracy. Also, using algorithms that are inherently better at handling imbalances, such as ensemble methods like Random Forest or XGBoost, can be beneficial.

Describe your experience with time series analysis.

I've worked extensively with time series analysis, primarily in the context of forecasting and anomaly detection. Using tools like Python's pandas and statsmodels, I've handled datasets involving sales figures, web traffic, and sensor data.

A memorable project involved forecasting monthly sales for a retail chain. I leveraged ARIMA and seasonal decomposition techniques to model the inherent seasonality in the data, allowing the business to better manage inventory and staffing levels. Additionally, I’ve used specialized libraries like Facebook Prophet for its ability to handle holidays and other irregularities with ease.

What are the key steps in a data analysis project?

A data analysis project typically starts with defining the problem or question you're trying to answer, which informs your objectives and the scope of the analysis. Next, you'll collect and clean the relevant data to ensure it's reliable and suitable for analysis. This often involves dealing with missing values, outliers, or inconsistent data formats.

After preprocessing, you explore the data through descriptive statistics and visualization to identify patterns, trends, or anomalies. Based on these insights, you might apply more advanced analytical techniques or build models to draw deeper conclusions. Finally, you'll interpret the results, relate them back to your initial objectives, and communicate your findings in a clear and actionable manner, often through a combination of reports, dashboards, or presentations.

How do you prioritize tasks when working on multiple analytics projects simultaneously?

When handling multiple analytics projects, the first thing I do is understand the priorities and deadlines of each project. I'll typically start by discussing with stakeholders to get clarity on what's most critical for the business. Once I have that information, I use a mix of project management tools and techniques like the Eisenhower Matrix to categorize tasks by urgency and importance.

I also break down each project into smaller, manageable tasks and set mini-deadlines to keep progress on track. Regularly revisiting and adjusting priorities based on any new information or shifting deadlines helps ensure that I'm always working on the most impactful tasks. Good communication with the team and stakeholders is crucial to keep everyone aligned and manage expectations effectively.

Explain how you would determine the reliability of your data source.

I'd start by assessing the data source's reputation and track record. Are they well-known in the industry for providing accurate and timely data? Next, I'd look into the methods they've used for data collection and processing to ensure they conform to best practices and standards. Additionally, I would check their data for consistency and completeness over time; any significant anomalies could indicate issues. Lastly, I'd validate the data against other trusted sources to see if they align well.

How do you handle large datasets that cannot fit into memory?

When dealing with large datasets that can't fit into memory, I typically use a combination of techniques like data sampling, chunk processing, and leveraging tools designed for big data, such as Apache Spark or Dask. Data sampling can help get quick insights without needing the entire dataset. For processing, I might use chunking, breaking the dataset into smaller pieces, and processing them sequentially or in parallel. Tools like Spark allow for distributed processing, handling datasets across multiple nodes, which is efficient for large-scale data operations.

What is the use of random forest in data analysis?

Random forest is great for handling both classification and regression tasks in data analysis. It's an ensemble method that builds multiple decision trees and merges them to get a more accurate and stable prediction. This approach helps to reduce overfitting and improve the generalization of the model.

Another advantage is its ability to handle large datasets with higher dimensionality. It can manage missing values and maintain accuracy even when a large portion of the data is missing. Plus, random forests also provide insights into feature importance, which can be valuable for understanding the key drivers of your model's predictions.

Explain the difference between a histogram and a bar chart.

A histogram and a bar chart look similar but serve different purposes. A histogram displays the distribution of numerical data and groups it into bins or intervals. The y-axis represents the frequency or count of data points within each bin, making it ideal for showing the shape of data distribution over continuous intervals.

On the other hand, a bar chart compares categorical data where each bar represents a distinct category. The x-axis represents different categories, while the y-axis shows the value associated with each category, such as counts or percentages. So, while histograms give you insight into how data is distributed across a range, bar charts help you compare different categories directly.

How do you perform data cleaning before analysis?

Data cleaning involves identifying and correcting inaccuracies or inconsistencies in your dataset. First, I check for missing values and decide on the best way to handle them—either by filling them in, if there’s a logical value to replace them with, or by excluding rows or columns with too many missing values. Next, I look for duplicates and ensure they're handled appropriately, often by removing them if they're truly redundant. I also validate the data types to ensure they are appropriate for the analysis, like making sure dates are in date format and numeric values aren't stored as text. Finally, I perform sanity checks to confirm that the data ranges and distributions make sense and don't contain outliers or impossible values that could skew the analysis.

Describe the concept of normalization and why it is important in data analysis.

Normalization is a process used to organize a database into tables and columns to minimize redundancy and dependency. It involves splitting large tables into smaller, more manageable pieces and defining relationships between them. This ensures that each piece of data is stored only once, improving data integrity and reducing anomalies.

Normalization is crucial because it makes databases more efficient and easier to maintain. By eliminating redundant data, you save storage space and reduce the risk of inconsistencies. It also enhances query performance because normalized tables tend to be simpler and more focused, allowing for quicker retrieval and updates. This approach makes data analysis more accurate and reliable, as you’re working with clean, well-organized datasets.

What metrics would you use to measure the performance of a model?

The key metrics often depend on the type of model and the problem you're tackling, but generally, for a classification model, you'd look at accuracy, precision, recall, and F1 score. If it's a regression model, you'd focus on metrics like Mean Absolute Error (MAE), Mean Squared Error (MSE), and R-squared.

For classification, precision and recall help you understand the trade-off between false positives and false negatives. For regression, MAE gives an idea of the average magnitude of errors in a set of predictions, while MSE penalizes larger errors more heavily. R-squared tells you how well the model explains the variability of the target variable.

In addition, you might consider AUC-ROC for classification tasks to evaluate the trade-off between true positive rate and false positive rate across different threshold settings. Also, cross-validation is crucial to ensure that your metrics are consistent and the model generalizes well to new data.

Discuss a time when your analysis was incorrect

There was a time when I was working on a sales forecast model for a retail company. I underestimated the impact of a major promotional event because I didn’t account for the historical data from similar events in my model. As a result, my forecast was overly conservative and didn’t predict the sales surge accurately.

When the actual sales numbers came in, they were significantly higher than my projections. I had to quickly analyze where I went wrong and present the revised model to the stakeholders. I learned a valuable lesson about the importance of incorporating all relevant variables and the need for rigorous sensitivity analysis in predictive modeling.

How do you decide which variables to include in a regression model?

Choosing variables for a regression model often starts with understanding the domain and the problem you're trying to solve. You typically begin by including variables that you believe, based on prior knowledge or theory, could influence the outcome. Then you can use statistical methods like correlation analysis to see which variables are strongly associated with the dependent variable.

Another approach is to use techniques such as stepwise regression, which systematically adds or removes variables based on their statistical significance. Additionally, considering multicollinearity is important; you can use Variance Inflation Factor (VIF) to check and avoid highly correlated predictors. Ultimately, balancing model complexity and interpretability will guide your decision on which variables to keep.

What is clustering and what are its types?

Clustering is an unsupervised machine learning technique used to group similar data points together based on certain characteristics or features. The goal is to make sure that objects in the same group (or cluster) are more similar to each other than to those in other groups.

There are several types of clustering, including:

  1. K-Means Clustering: One of the most popular methods, where the dataset is divided into K clusters, and each data point belongs to the cluster with the nearest mean.
  2. Hierarchical Clustering: This can be either agglomerative (bottom-up) or divisive (top-down). It builds a tree-like structure of the data in clusters.
  3. DBSCAN (Density-Based Spatial Clustering of Applications with Noise): This algorithm groups together closely packed points and marks points that are in low-density regions (noise).
  4. Mean Shift Clustering: A centroid-based algorithm that updates candidates for centroids to be the mean of the points within a given region.

Explain the importance of data visualization in analytics.

Data visualization is crucial in analytics because it transforms complex data sets into intuitive and interactive graphs, charts, and maps that make insights accessible and comprehensible. It helps to quickly identify trends, patterns, and outliers, which would be difficult to detect in raw data form.

Moreover, data visualization aids in better communication of findings to stakeholders who may not be data-savvy. By presenting data visually, you can tell compelling stories that drive informed decision-making and insights, making it easier to achieve buy-in on strategic decisions.

Explain the difference between ROC and AUC curves.

ROC, or Receiver Operating Characteristic curve, is a graphical representation of a classifier's performance. It plots the True Positive Rate (sensitivity) against the False Positive Rate (1-specificity) at various threshold settings. Essentially, it shows how well your model can distinguish between classes.

AUC, or Area Under the Curve, represents the degree or measure of separability achieved by the model. It quantifies the entire area under the ROC curve, providing a single scalar value that summarizes the performance of the model. AUC can range from 0 to 1, where 1 indicates perfect separability and 0.5 suggests no better than random guessing.

Explain the concept of the Null Hypothesis.

The Null Hypothesis is a fundamental concept in statistical analysis. It essentially states that there is no effect or no difference, and it serves as a starting point for any statistical test. For example, if you're testing a new drug, the null hypothesis would claim that the drug has no effect compared to a placebo. You then use data and statistical tests to try and provide evidence against this hypothesis, aiming to reject it in favor of an alternative hypothesis that suggests a real effect or difference exists.

How would you present the results of your analysis to a client?

I'd focus on clarity and relevance. Start with a clear and concise executive summary that highlights the key findings and their implications. Use visual aids like charts, graphs, and dashboards to make complex data more digestible. Tailor the presentation to the client's level of expertise, ensuring that technical jargon is kept to a minimum or properly explained. Provide actionable insights and recommendations that address the client's specific business goals, making it clear what steps they should take next based on the analysis.

What is multicollinearity and how can it be detected?

Multicollinearity occurs when two or more independent variables in a regression model are highly correlated, making it difficult to distinguish their individual effects on the dependent variable. This can inflate the variance of the coefficient estimates and make the model less reliable.

To detect multicollinearity, you can look at the Variance Inflation Factor (VIF) for each predictor variable. A VIF value greater than 10 often indicates significant multicollinearity. Additionally, examining the correlation matrix of the predictor variables can help identify pairs of highly correlated variables. If two variables have a correlation coefficient close to ±1, that's another signal for multicollinearity.

How do you perform feature selection in a dataset?

Feature selection can be approached in several ways, depending on the context and the size of the dataset. Common methods include using statistical tests like chi-square for categorical variables or correlation coefficients for numerical ones to determine the most significant features. Regularization techniques like Lasso (L1 regularization) also help by penalizing less important features more heavily, effectively shrinking their coefficients to zero. Another practical method is using tree-based algorithms like Random Forests, which provide feature importance scores based on how often and effectively features are used to split the data.

Describe logistic regression and its applications.

Logistic regression is a statistical method used for binary classification problems, where the outcome is a binary variable that can take on two possible outcomes, usually coded as 0 or 1. It models the probability that a given input point belongs to a certain class. Essentially, it uses the logistic function, also known as the sigmoid function, to map predicted values to probabilities.

Applications are widespread. In marketing, it can be used to predict whether a customer will purchase a product or not. In healthcare, it is used for diagnosing diseases based on symptoms. In finance, logistic regression helps in credit scoring to determine the probability of a borrower defaulting on a loan. The method is favored for its simplicity and interpretability.

What is a confusion matrix and how is it used?

A confusion matrix is a table that is used to evaluate the performance of a classification algorithm. It displays the true positives, true negatives, false positives, and false negatives in a matrix format, allowing you to see how many predictions were correctly classified versus how many were incorrectly classified. This helps in understanding the accuracy, precision, recall, and F1 score of your model, giving a more nuanced view than just the overall accuracy rate.

For example, in a binary classification problem, the confusion matrix shows you the number of correct predictions for each class and the number of incorrect predictions. This enables you to identify if your model is better at predicting one class over another and whether there's an imbalance or a specific type of error that is occurring more frequently.

Get specialized training for your next Analytics interview

There is no better source of knowledge and motivation than having a personal mentor. Support your interview preparation with a mentor who has been there and done that. Our mentors are top professionals from the best companies in the world.

Hi, my name is Tristan. I am a data evangelist! I am the former Global Head of Data and Analytics for Pizza Hut's Digital Business outside of the US. Prior to that, I've worked in data roles within investment banking as well as start ups. I am extremely passionate about …

$240 / month
2 x Calls

Only 1 Spot Left

Hi, I'm a seasoned Analytics professional with over a decade of hands-on experience in Deloitte Consulting and Google Inc. I'm passionate about using data-driven insights to solve complex business problems and guide strategic decision-making. [Growth in Tech] After achieving success with 4 promotions in BigTech over the last 10 years, …

$80 / month
2 x Calls

Only 4 Spots Left

Seasoned Strategy and Growth professional spearheading a $200 M+ category at Uber. I have 2.5+ years of mentorship experience and have helped my mentees crack jobs at top firms including EY Parthenon, Uber, Cadbury, Swiggy, as well as helped them reach top Business schools.

$90 / month
Regular Calls

Only 1 Spot Left

As a mentor with a background in both research and industry, I have a wealth of experience of 10+ years to draw upon when guiding individuals through the field of machine learning. My focus is on helping experienced software engineers transition into ML/DS, as well as assisting machine learning engineers …

$150 / month
Regular Calls

Only 2 Spots Left

Are you ready to take your product management career to the next level? I have helped 20+ mentees developing their careers, help get into product management, and assist entrepreneurs in launching their products. 🔍 Finding product market fit, using various discovery processes 📈 Advancing your own career in product management …

$180 / month
1 x Call

Only 2 Spots Left

Hi, I am a data scientist with more than 10 years of experience in the field of data science covering a wide range of fields and methods, such statistics, machine learning, optimisation, network analysis, time-series forecasting, agent based modelling for understanding the complexities of consumer behaviour, geo-spatial data science and …

$210 / month
4 x Calls

Browse all Analytics mentors

Still not convinced?
Don’t just take our word for it

We’ve already delivered 1-on-1 mentorship to thousands of students, professionals, managers and executives. Even better, they’ve left an average rating of 4.9 out of 5 for our mentors.

Find a Analytics mentor
  • "Naz is an amazing person and a wonderful mentor. She is supportive and knowledgeable with extensive practical experience. Having been a manager at Netflix, she also knows a ton about working with teams at scale. Highly recommended."

  • "Brandon has been supporting me with a software engineering job hunt and has provided amazing value with his industry knowledge, tips unique to my situation and support as I prepared for my interviews and applications."

  • "Sandrina helped me improve as an engineer. Looking back, I took a huge step, beyond my expectations."