Adept AI India Data Scientist Interview: Rounds, Questions and Preparation

What the job involves

Data Scientists in AI and deeptech work on building and deploying machine learning models that power products such as large language models, recommendation systems, or computer vision pipelines. Day to day, they clean and analyze large datasets, run experiments, train and evaluate models, and collaborate with engineers to ship models to production. At fast-growing companies like Adept AI India, they often wear multiple hats, from data collection to model monitoring.

Role and Technical Knowledge

1. 1

   Explain the bias-variance tradeoff and how you would diagnose it in a model you built.

   A strong answer defines bias and variance, explains underfitting vs overfitting, and mentions techniques like learning curves, cross-validation, and regularization to diagnose and address the tradeoff.

2. 2

   How would you handle missing data in a dataset for training an AI model?

   Cover different types of missingness (MCAR, MAR, MNAR), simple imputation (mean, median), advanced methods (KNN imputation, multiple imputation), and when to drop rows or columns. Mention that the choice depends on the problem and data.

3. 3

   What is the difference between L1 and L2 regularization? When would you use each?

   Explain that L1 (Lasso) adds absolute weights and can lead to sparse solutions, while L2 (Ridge) adds squared weights and shrinks coefficients. Use L1 for feature selection, L2 when all features are useful but correlated.

4. 4

   Describe how a transformer model works at a high level. Why is it important in modern NLP?

   Mention self-attention, positional encodings, encoder-decoder architecture, and how it enables parallel processing and capturing long-range dependencies. Highlight its role in models like BERT and GPT.

Practical Scenarios

1. 5

   You are given a dataset of customer reviews for a new AI product. How would you build a sentiment analysis model from scratch?

   Outline steps: data cleaning, text preprocessing (tokenization, stopword removal), feature extraction (TF-IDF or embeddings), model selection (logistic regression, LSTM, or fine-tuning a transformer), evaluation (accuracy, F1), and deployment considerations.

2. 6

   Your model performs well on validation but poorly on test data. What could be the reasons and how would you fix it?

   Discuss overfitting, data leakage, distribution shift, or improper validation split. Suggest solutions like cross-validation, regularization, collecting more data, or using domain adaptation.

3. 7

   Imagine you need to deploy a model to production for real-time predictions. What steps would you take?

   Cover model serialization (pickle, ONNX), building an API (Flask, FastAPI), containerization (Docker), cloud deployment (AWS SageMaker, GCP AI Platform), and monitoring for drift and performance.

4. 8

   How would you explain a complex model's predictions to a non-technical stakeholder?

   Mention techniques like SHAP, LIME, or feature importance. Emphasize using simple visualizations and focusing on actionable insights rather than technical details.

Behavioural and HR

1. 9

   Tell me about a time you worked on a team project. What was your role and how did you handle disagreements?

   Use the STAR method. Highlight collaboration, communication, and how you resolved conflicts constructively. Show that you value team success over individual credit.

2. 10

   Why do you want to work at a fast-growing AI startup like Adept AI India rather than a large tech company?

   Express enthusiasm for impact, learning, and wearing multiple hats. Mention specific aspects of the company's work (e.g., large language models, AI agents) and how it aligns with your career goals.

3. 11

   Describe a situation where you had to learn a new technology or tool quickly to complete a task.

   Show adaptability and self-learning. Give a concrete example, such as picking up PyTorch for a project, and explain how you approached learning and applied it successfully.

4. 12

   Where do you see yourself in 3-5 years in the field of AI and deeptech?

   Align your growth with the company's trajectory. Mention deepening expertise in areas like deep learning or MLOps, taking on more responsibility, and contributing to impactful AI products.

Coding and Problem Solving

1. 13

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

   Implement precision and recall from confusion matrix components, then compute F1 as 2*(precision*recall)/(precision+recall). Handle division by zero gracefully.

2. 14

   Given a list of integers, find the longest increasing subsequence. How would you optimize it?

   Start with a dynamic programming O(n^2) solution, then mention the O(n log n) approach using binary search. Explain the intuition and trade-offs.

3. 15

   Write a SQL query to find the top 3 most frequent words in a table of customer feedback.

   Use string splitting functions (e.g., SPLIT_PART or REGEXP_SPLIT_TO_TABLE), then GROUP BY word, ORDER BY count DESC, and LIMIT 3. Discuss handling case sensitivity and punctuation.

4. 16

   How would you detect and handle outliers in a dataset before training a model?

   Mention methods like Z-score, IQR, or isolation forests. Discuss whether to remove, cap, or transform outliers, and the impact on model performance.