ML & LLM Learning: Coding Interview Preparation

A comprehensive repository for ML and LLM coding interview preparation with implementations, theory, and interview Q&A.

🔥 Frontier-Lab Interview Prep — Start Here

These deep-dive and grill files are the highest-yield content in the repo for applied scientist / ML engineer interviews at frontier labs and big tech. Each topic has a *_DEEP_DIVE.md (interview-grade theory) plus an INTERVIEW_GRILL.md (50–60 brutal questions with strong answers). Drill the grill files until you can answer 40+ cold.

Recommended drill sequence: (1) Read each *_DEEP_DIVE.md start to finish. (2) Drill INTERVIEW_GRILL.md until 40+/60 cold. (3) Cycle back through the misses the next day. (4) Ask a friend to randomly pick 10 questions per topic and grill you out loud.

The remaining content of this repo (60+ topic folders) is supporting material. The five interview-grade pairs above are the highest-leverage files.

🎯 What You'll Learn

This repository covers everything you need for ML/LLM coding interviews:

• Classical ML Algorithms - Simple implementations (pure Python/NumPy + PyTorch)
• Evaluation Metrics - All common metrics with simple code
• Transformers & Attention - Core concepts with simple implementations
• LLM Inference Techniques - KV cache, quantization (simple code)
• Attention Mechanisms - Different types with clear code
• LLM Problem Solving - Long context, efficiency solutions
• Training Techniques - RLHF, DPO (simplified implementations)
• Sampling Techniques - Top-p, nucleus, temperature (pure Python)
• Optimizers - SGD, Adam, etc. (from scratch)
• Regularization - L1, L2, dropout (simple implementations)
• Theory & Interview Q&A - Comprehensive coverage
• Diffusion Models - Complete theory, training, evaluation, NLP applications
• Mixture of Experts (MoE) - Architecture, routing, load balancing, efficiency
• State Space Models (SSM) - Mamba, linear complexity, long sequence modeling
• Language Modeling Losses - MLM, CLM, NSP implementations and explanations
• Normalization Techniques - BatchNorm and LayerNorm with detailed theory and implementations
• Reinforcement Learning Fundamentals - MDP, Q-Learning, Multi-Armed Bandit, Monte Carlo in easy language
• RNN and LSTM - Simple, short, precise implementations from scratch

💡 Code Philosophy

All code is kept simple:

• Pure Python/NumPy versions - No heavy dependencies, easy to understand
• PyTorch versions - Simple PyTorch implementations for comparison
• From scratch - Understand how things work internally
• Interview-ready - Code you can write in interviews

📁 Repository Structure

ml_and_llm_learning/
├── 00_pytorch_fundamentals/      # PyTorch basics (START HERE if new to PyTorch)
├── 01_classical_ml/              # Linear/logistic regression, KNN, K-means
├── 02_gradient_descent/          # Different GD variants
├── 03_evaluation_metrics/        # All evaluation metrics
├── 04_transformers/              # Transformer architecture
├── 05_attention_mechanisms/      # Different attention types
├── 06_llm_inference/             # KV cache, optimization
├── 07_llm_problems/              # Long context, efficiency
├── 08_training_techniques/       # RLHF, DPO, PPO, GRPO
├── 09_sampling_techniques/       # Top-p, nucleus, temperature
├── 10_optimizers/                # Optimizer implementations
├── 11_regularization/            # Regularization techniques
├── 12_theory/                    # Comprehensive theory
├── 13_interview_qa/              # Interview questions & answers
├── ...
├── 47_statistical_inference/     # Estimators, CIs, tests, Bayesian updates
├── 48_optimization_and_matrix_calculus/  # Gradients, Hessians, conditioning
├── 49_generalization_and_evaluation/     # Leakage, calibration, ablations
├── 50_ml_coding_interview_patterns/      # Pressure-friendly coding templates
├── 51_llm_research_interview_prep/       # LLM eval, ablations, research judgment
├── ...
├── 62_frontier_training_playbook/        # Architecture, stability, data, ablations
├── 63_paged_attention_and_llm_serving/   # KV cache, fragmentation, paging, batching
└── 64_integrated_ai_ml_interview_synthesis/  # Cross-topic interview answer patterns

🚀 Quick Start

1. Set Up Environment

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate
pip install -r requirements.txt

2. Learn PyTorch Basics (If Needed)

# If you're new to PyTorch, start here
cd 00_pytorch_fundamentals
python pytorch_basics.py

3. Start Learning

# Classical ML
cd 01_classical_ml
python linear_regression.py

# Transformers
cd 04_transformers
python attention.py

# LLM Inference
cd 06_llm_inference
python kv_cache.py

📚 Learning Path

See LEARNING_PATH.md for the complete learning journey.

🎓 Prerequisites

• Python 3.9+
• Basic Python knowledge
• Understanding of linear algebra
• (Optional) PyTorch/TensorFlow experience

🔧 Technologies

• NumPy: Numerical computations (pure Python/NumPy implementations)
• PyTorch: Simple PyTorch versions (optional, for comparison)
• Matplotlib: Visualization (optional)
• Pure Python: Most code uses only NumPy (minimal dependencies)

📖 Topics Covered

1. Classical ML - Linear/logistic regression, KNN, K-means
2. Gradient Descent - Batch, SGD, Mini-batch, Adam
3. Evaluation Metrics - Accuracy, precision, recall, F1, etc.
4. Transformers - Architecture, attention, decoding
5. Attention Mechanisms - Self-attention, cross-attention, etc.
6. LLM Inference - KV cache, quantization, optimization
7. LLM Problems - Long context, efficiency solutions
8. Training Techniques - RLHF, DPO, PPO, GRPO
9. Sampling Techniques - Top-p, nucleus, temperature
10. Optimizers - SGD, Adam, AdamW, etc.
11. Regularization - L1, L2, dropout, etc.
12. Theory - Comprehensive ML/LLM theory
13. Interview Q&A - 100+ interview questions
14. Advanced Positional Embeddings - RoPE, ALiBi
15. Tokenization - BPE, WordPiece, SentencePiece
16. Training Behaviors - Single GPU, loss spikes
17. Probability Math - Common probability Q&A
18. Distribution Classification - Which distribution?
19. Advanced Clustering - Hierarchical, DBSCAN, GMM
20. Multi-Turn Conversations - Design & long context
21. Dimensionality Reduction - PCA, theory & math
22. Recommendation Systems - Matrix factorization, evaluation
23. Clustering Evaluation - Silhouette, ARI, NMI
24. Linear Algebra Q&A - Eigenvalues, SVD, rank
25. Adapters & LoRA - Parameter-efficient fine-tuning
26. Tree-Based Methods - Decision Tree, Random Forest, Gradient Boosting, XGBoost
27. Advanced Theory - Bias-variance, cross-validation, learning curves
28. Business Use Cases - Churn, recommendations, fraud, pricing (detailed solutions)
29. System Design for ML - Scalable pipelines, serving, monitoring
30. A/B Testing - Statistical testing, sample size, interpretation
31. Neural Networks - Forward pass, backpropagation from scratch (detailed)
32. Anomaly Detection - Isolation Forest (detailed explanation, when to use)
33. Information Theory - Entropy, KL divergence, cross-entropy, mutual information, Gini
34. Discriminative vs Generative - Model types, assumptions (Linear, Logistic, SVM), Bayes' theorem
35. Kernel Functions - Linear, Polynomial, RBF, Sigmoid (detailed explanations, when to use)
36. NLP Basics - TF-IDF, N-grams, Laplace smoothing, L1/L2 priors (detailed explanations)
37. MLE and MAP Estimation - Maximum Likelihood, Maximum A Posteriori (detailed derivations)
38. Multimodal Models & Embedding History - CLIP, embedding training, NLP evolution (TF-IDF → Word2Vec → GloVe → BERT)
39. RAG (Retrieval-Augmented Generation) - Industry-standard architecture, challenges, solutions, evaluation (production-ready)
40. Diffusion Models - Complete theory, training, evaluation, NLP applications
41. Mixture of Experts (MoE) - Architecture, routing, load balancing, efficiency
42. State Space Models (SSM) - Mamba, linear complexity, long sequence modeling
43. Language Modeling Losses - MLM, CLM, NSP implementations and explanations
44. Normalization Techniques - BatchNorm and LayerNorm with detailed theory and implementations
45. Reinforcement Learning Fundamentals - MDP, Q-Learning, Multi-Armed Bandit, Monte Carlo in easy language
46. RNN and LSTM - Simple, short, precise implementations from scratch
47. Statistical Inference - Estimators, MLE, confidence intervals, hypothesis tests, bootstrap, Bayesian updates
48. Optimization and Matrix Calculus - Gradients, Jacobians, Hessians, convexity, conditioning, optimizer intuition
49. Generalization and Evaluation - Leakage, calibration, class imbalance, distribution shift, ablations, metric uncertainty
50. ML Coding Interview Patterns - Stable softmax, masking, vectorization, top-k/top-p, padding, k-means update templates
51. LLM Research Interview Prep - Perplexity, pass@k, retrieval metrics, ablation reasoning, paper discussion structure
52. Statistical Learning Theory - Empirical vs population risk, capacity, generalization gap, regularization as inductive bias
53. ML Debugging and Mock Coding - Timed coding prompts, NaN/debugging patterns, leakage checks, training failure diagnosis
54. Data Manipulation for ML - Pandas feature-table work, joins, groupby, normalization, preprocessing without leakage
55. Research Papers and Mock Interviews - Paper discussion prompts, research judgment, and probability questions like distribution membership
56. Spoken Interview Question Bank - Live-interview style model answers for ML theory, coding, probability, and LLM research questions
57. Meta-Style Mock Interviews - Full simulated technical loops, follow-ups, and scoring rubric
58. Whiteboard Derivations - Stepwise must-master derivations and memory skeletons
59. Blind Coding Drills - No-search, timed implementation drills from memory
60. Research Judgment Rounds - Scenario-based evidence review, ablations, and claim evaluation
61. Large-Scale LLM Systems - Training-memory, sharding, parallelism, serving, and scale trade-offs
62. Frontier Training Playbook - Methodology for architecture choices, stability, data mixture, and believable ablations
63. Paged Attention and LLM Serving Internals - KV-cache fragmentation, block tables, prefix sharing, continuous batching, and serving bottlenecks
64. Integrated AI and ML Interview Synthesis - Bridges across theory, coding, systems, and research judgment with answer frameworks

🎯 Prerequisites

New to PyTorch? Start with 00_pytorch_fundamentals/ to learn all PyTorch concepts you'll need for this repository.

🎯 Learning Goals

By completing this repository, you'll be able to:

• ✅ Implement ML algorithms from scratch
• ✅ Understand transformer architecture deeply
• ✅ Optimize LLM inference
• ✅ Answer interview questions confidently
• ✅ Understand training techniques (RLHF, DPO, etc.)
• ✅ Implement evaluation metrics
• ✅ Understand theory behind ML/LLM

Ready to start? Open LEARNING_PATH.md and begin your journey! 🚀