Microsoft Applied Scientist (AI/ML) New Grad Interview 2024 | Hyderabad

Round 1: Online Assessment

• Duration: 90 mins
• Platform: Codility
• Questions: 2 coding questions (One Array manipulation, One Graph BFS) and 15 MCQs based on Machine Learning fundamentals (Loss functions, Gradient Descent, Overfitting).

Round 2: ML Fundamentals & Coding

• Duration: 60 mins
• Questions:
  1. Implement a K-Means Clustering algorithm from scratch in Python.
  2. Given a scenario of highly imbalanced fraud detection data, how would you design the evaluation metrics?
• How to Answer: Do not just say 'Accuracy'. For imbalanced datasets, discuss Precision, Recall, F1-Score, and AUC-ROC. When coding algorithms from scratch, use NumPy for vectorization instead of writing basic for loops, which shows you understand computational efficiency.

Round 3: Deep Learning & AI System Design

• Duration: 60 mins
• Questions: Deep dive into my college NLP project. The interviewer asked to explain the mathematical difference between RNNs, LSTMs, and the self-attention mechanism in Transformers.
• How to Answer: For AI roles, math is just as important as code. I drew the architecture of a Transformer on the whiteboard and explained how Query, Key, and Value matrices are multiplied. If you don't know the exact math, admit it but explain the high-level intuition (e.g., 'Attention helps the model remember long-term context better than LSTMs by looking at the entire sequence at once').

Round 4: Hiring Manager

• Duration: 45 mins
• Questions: Discussion on Large Language Models (LLMs), fine-tuning vs RAG (Retrieval-Augmented Generation), and behavioral questions about handling failing models.

Key Takeaways for Freshers

For Applied Scientist roles, your theoretical knowledge of why an algorithm works is tested heavily. Don't just learn to use PyTorch or scikit-learn; understand the underlying mathematics of loss functions and optimizers.