• Deep learning discussion, Focused on understanding of core concepts.
• Explain how Neural Network works ?
• Explain different regularization techniques, how dropout regularization works ?
• Why do we need relu activation function?
• BERT: Pre-training and fine tuning strategies discussed in depth.
• Expalin trnasformer architecture? what is positional encoding why is it needed
• what is the objective function used in the BERT pre training ?
• how will you create your dataset for BERT pre-training, how will you tokenize your dataset ?
• Current trends and architecture in large language models (LLMs).

Onsite Round 1 : Machine Learning Discussion

• Detailed deep dive into BERT Architecture.
• Read BERT Paper
• Topics covered:
  • Model pre-training
  • Model fine-tuning
  • Transformer architecture internals
• Classical Machine Learning: Concepts around XGBoost.
• Explain how Gradient Boosting works
• How boosting is different from bagging
• what is the role of learning rate , explain it mathematically
• What is tree pruning , Explain cost complexity Tree pruning method
• Implementation of Beam Search algorithm. Write a python code to implement beam search algorithm

Onsite Round 2 : Coding DSA

• Problem: Delete a given node in a linked list without access to the head pointer.
• Discussion around:
  • Stack and queue implementation
  • Operating Systems: Concept of semaphores

Onsite Round 3 : System Design + Coding (DSA)

• ML System Design (Half of the round):
  • Design a spam filter for Apple phones with on-device deployment.
  • Design a QA system to answer history-related questions.
• Coding Question: Search in a rotated sorted array (with duplicates).

Onsite Round 4 : Coding (DSA Extra Round)

• Rotate Linked List
• 3 Sum Problem
• Number of Islands

Preparation Tips

• Prepare DSA from Easy to Medium level, focus on speed and accuracy.
• Customize your ML prep based on the team's work. Since the team was focused on text-heavy work (Apple Intellegence), Hence Most of my discussions were around BERT , model pre training , fine tuning etc.