Mad Street Den Machine Learning Engineer Interview: Rounds, Questions and Preparation

What the job involves

Machine Learning Engineers in AI and deeptech build and deploy models that power products such as visual search, recommendation systems, and document intelligence. Day to day, they clean data, train and evaluate models, optimise inference for latency and cost, and work with backend teams to ship models to production. Freshers typically start with well-scoped model iterations and data pipeline tasks under senior guidance.

Role and technical knowledge

1. 1

   Explain the bias-variance tradeoff and how you would detect and address underfitting or overfitting in a model.

   Define bias and variance clearly, mention learning curves, and give practical fixes like more data, regularisation, or model complexity changes.

2. 2

   What is the difference between bagging and boosting? Give an example algorithm for each.

   Explain that bagging trains models in parallel on bootstrapped data (Random Forest) while boosting trains sequentially to correct errors (XGBoost, AdaBoost).

3. 3

   How does a convolutional neural network work for image classification? Mention convolution, pooling, and fully connected layers.

   Describe how filters extract local features, pooling reduces spatial size, and the final layers map features to class scores. Mention parameter sharing and translation invariance.

4. 4

   You have a dataset with 95% negative and 5% positive examples. Which metrics would you use and why?

   Avoid accuracy. Discuss precision, recall, F1, and PR-AUC, and explain how class weights or resampling can help.

Practical scenarios and problem solving

1. 5

   Your model performs well on validation but poorly in production. How would you debug this?

   Check for data drift, train-serve skew, leakage, and differences in preprocessing. Suggest monitoring and retraining strategies.

2. 6

   You need to reduce inference latency of an image model by half without retraining from scratch. What options do you have?

   Discuss quantisation, pruning, knowledge distillation, smaller architectures, batching, and using ONNX or TensorRT.

3. 7

   How would you build a visual search feature that finds similar products from an image? Outline the pipeline.

   Cover embedding extraction with a CNN or transformer, indexing with FAISS or similar, and a ranking step. Mention evaluation with recall@k.

4. 8

   A model you deployed starts giving biased predictions for a certain user group. How do you investigate and mitigate?

   Talk about slicing metrics by group, checking data representation, and using techniques like reweighting, augmentation, or fairness constraints.

Coding and implementation

1. 9

   Write a Python function to compute the F1 score from scratch given true and predicted labels.

   Show clear handling of edge cases like zero division and explain each step. Use simple loops or NumPy.

2. 10

   Implement a function that splits a dataset into train and validation sets while preserving class proportions.

   Use stratified sampling. Explain why random split can fail for imbalanced data.

3. 11

   Given a list of model predictions and true labels, write code to plot a confusion matrix.

   Use sklearn's confusion_matrix or build it manually, then plot with matplotlib. Label axes and explain what each cell means.

4. 12

   How would you write a PyTorch training loop for a simple classifier? Outline the key steps.

   Mention DataLoader, model.train(), zero_grad, forward pass, loss, backward, optimizer step, and validation with model.eval().

Behavioural and HR

1. 13

   Tell me about a project where you had to learn a new ML technique quickly. How did you approach it?

   Use STAR format. Focus on your learning process, resources used, and the outcome. Show curiosity and self-drive.

2. 14

   Describe a time you worked in a team and disagreed with a teammate's modelling approach. What did you do?

   Show respect, data-driven discussion, and willingness to test both ideas. End with a positive resolution.

3. 15

   Why do you want to work in AI and deeptech, and why this role specifically?

   Connect your interest to real products like visual AI or recommendation systems. Mention the company's domain and how your skills fit.

4. 16

   Where do you see yourself in three years as an ML engineer?

   Show a growth path from executing tasks to owning models and mentoring. Align with the company's need for end-to-end ML ownership.