SQL Questions and Answers for Business Analyst (BA) Role
🔹 Experience-Based Questions
📌 Celebal Technologies – Data Scientist Intern
Q1. What business problem were you solving with the churn prediction model? We aimed to identify
customers likely to leave the company so that proactive retention strategies could be applied. By predicting
churn, we helped the business minimize customer loss and improve customer satisfaction.
Q2. Why did you choose Random Forest for churn prediction? Did you compare it with other models?
Random Forest was chosen because it is robust, handles both categorical and numerical data, and provides
good performance out-of-the-box. Yes, I compared it with Logistic Regression and Decision Trees. Random
Forest gave the best accuracy and stability, and also offered useful feature importance scores.
Q3. How did you use GridSearchCV and what parameters did you tune? I used GridSearchCV to fine-
tune hyperparameters like n_estimators , max_depth , and min_samples_split . It helped
automate the parameter tuning process and improved model accuracy during cross-validation.
Q4. Why was SMOTE necessary in your case, and what metrics improved after using it? The dataset
was imbalanced, with far fewer churned customers. SMOTE helped synthetically balance the classes. As a
result, recall and F1-score for the minority class (churned users) improved significantly without sacrificing
overall accuracy.
Q5. Explain how you built the Streamlit dashboard. What insights did it offer to stakeholders? The
dashboard was built using Streamlit to make the model interactive and user-friendly. It allowed users to
input customer data and visualize churn risk instantly. I used charts to show churn distribution, feature
importance, and customer-level predictions. Stakeholders could understand which factors led to churn and
take preventive measures.
Q6. What do you mean by "real-time single-customer inference"? How was latency managed? It refers
to the model's ability to process one customer's data and return a prediction instantly. I optimized the
preprocessing pipeline (encoding, scaling) and used Joblib to load the model efficiently, ensuring low
latency.
Q7. How did you ensure reproducibility with Joblib? I saved the trained model, scaler, and encoders using
Joblib. During deployment or testing, I loaded them as-is, ensuring consistent predictions every time
without retraining.
📌 Bacancy Technologies – Software Developer Intern
Q8. How did you design your Django-based budget tracker? What were the key modules? I structured
the app using Django’s MVC architecture. Key modules included user authentication, budget entry forms,
1
category management, and expense summary views. The app allowed users to input daily budgets and
expenses, and view monthly summaries.
Q9. Can you walk me through the database schema and how you handled transactions? The schema
had tables for Users, Budgets, and Expenses. Each expense was linked to a category and user via foreign
keys. I used Django ORM for database operations and managed transactions implicitly using Django’s
atomic operations to ensure data consistency.
Q10. How did you ensure frontend responsiveness and user experience? I used HTML, CSS, and
Bootstrap to make the UI responsive. I ensured consistent layout, clear navigation, and used validation to
guide users during form submissions. The design was tested across browsers and devices for compatibility.
(The previous SQL section continues here... [unchanged])
