🎯 Interview Prep
A structured guide to passing ML and AI engineering interviews — from coding rounds to system design and behavioral questions.
Golden rule: Practice explaining concepts out loud. Interviewers evaluate clarity of thinking as much as correctness.
Interview Format
| Round | Focus | Duration | Prep Strategy |
|---|---|---|---|
| Screen | Resume + Background | 30 min | STAR method, highlight top 3 projects |
| Coding | DSA / Python | 45–60 min | NeetCode 75, pattern recognition |
| ML Theory | Concepts + Math | 45 min | StatQuest, review bias-variance, metrics |
| System Design | ML Architecture | 45–60 min | Practice recommender, search, serving designs |
| Take-home | End-to-end ML task | 4–8 hrs | Clean notebook, metrics, README |
| Final | Culture + Leadership | 30–45 min | Company values, impact stories |
ML Theory — Key Topics
Core ML Concepts
- Bias-variance tradeoff — when each matters
- Overfitting: causes, detection, prevention
- Precision vs Recall — when to optimize each
- ROC-AUC — what it measures and its limits
- Feature importance — model-agnostic vs model-specific
- Cross-validation — why k-fold beats train/val split
- Regularization — L1 (sparsity) vs L2 (smoothness)
Deep Learning Concepts
- Backpropagation — chain rule walkthrough
- Vanishing gradient — causes and solutions
- Batch norm — why it stabilizes training
- Attention — keys, queries, values intuition
- Transformer vs RNN — when to use each
- Transfer learning — freeze vs fine-tune decision
- Loss functions — CrossEntropy vs MSE vs contrastive
GenAI + LLMs
- RAG vs Fine-tuning — decision framework
- Hallucination — causes and mitigation
- Prompt injection — what it is, how to guard
- Token limits — chunking and retrieval tradeoffs
- LoRA/QLoRA — how PEFT reduces VRAM
- Embedding models — cosine similarity, dimensions
System Design — ML
- Design a recommendation system end-to-end
- Design a semantic search engine
- Design a real-time fraud detection system
- Model serving: latency vs throughput tradeoffs
- Data drift detection in production
- A/B testing models safely
DSA Patterns to Master
Must-Know Patterns
- Two Pointers — sorted arrays, pair sums
- Sliding Window — subarrays, longest substring
- Fast + Slow Pointers — cycle detection
- Hash Map counting — frequency, anagram problems
- BFS — level order, shortest path
- DFS — path finding, tree traversal
- Binary Search — on sorted arrays and answer space
- DP — 1D and 2D memoization
Recommended Practice
- NeetCode 150 — complete all Easy and Medium
- LeetCode 75 — the official curated study plan
- Solve by pattern, not by problem
- Time yourself: Easy <15 min, Medium <30 min
- Write complexity analysis for every solution