AIML Rounds
Track 01

Fundamentals

The core ideas behind every model, from maths and regression to neural nets.

Maths for ML

Module 1

Linear algebra

Vectors, matrices, eigenvectors and SVD.

  1. Vectors, Dot Products & Cosine SimilarityConcept20 min
  2. Matrices as TransformationsConcept20 min
  3. Eigenvalues & EigenvectorsConcept25 min
  4. SVD & Low-Rank ApproximationConcept25 min

Module 2

Calculus and optimisation

Gradients, the chain rule and gradient descent.

  1. Derivatives, Gradients & the Chain RuleConcept25 min
  2. Gradient Descent, Momentum & AdamConcept25 min
  3. Convexity & Local MinimaConcept20 min

Module 3

Probability

Bayes' rule, distributions, expectation and maximum likelihood.

  1. Probability Basics & Bayes' RuleConcept20 min
  2. Common Probability DistributionsConcept25 min
  3. Expectation, Variance & CovarianceConcept20 min
  4. Maximum Likelihood & MAPConcept25 min

Module 4

Probability practice

Interview questions on Bayes, coins, dice, cards and urns, with worked answers.

  1. Bayes' Rule PracticeInterview20 min
  2. Coin Flip PracticeInterview25 min
  3. Dice PracticeInterview20 min
  4. Playing Card PracticeInterview25 min
  5. Balls & Urns PracticeInterview20 min
  6. Classic Probability PuzzlesInterview25 min

Module 5

Statistics

Confidence intervals, hypothesis tests and A/B testing.

  1. Confidence Intervals & the BootstrapConcept25 min
  2. Hypothesis Tests, p-values & A/B TestingConcept30 min
  3. Correlation vs CausationConcept25 min

Module 6

Information theory

Entropy, cross-entropy and KL divergence.

  1. Entropy, Cross-Entropy & KL DivergenceConcept25 min

Classical ML

Module 7

Data and features

Turning raw columns into model-ready features.

  1. Encoding Categorical FeaturesConcept30 min
  2. Missing Values & Feature ScalingConcept30 min

Module 8

Linear models

Weighted sums of features: regression, regularisation and classification.

  1. Linear RegressionConcept30 min
  2. Regularized Linear ModelsConcept30 min
  3. Logistic RegressionConcept30 min
  4. Softmax RegressionConcept25 min

Module 9

Tree-based models

Splits, ensembles and boosting: the workhorses of tabular data.

  1. Decision TreesConcept30 min
  2. Random ForestConcept25 min
  3. Gradient BoostingConcept35 min

Module 10

Other classic classifiers

Margins, neighbours and probabilities.

  1. Support Vector MachinesConcept35 min
  2. k-Nearest NeighboursConcept30 min
  3. Naive BayesConcept30 min

Module 11

Unsupervised learning

Finding structure without labels.

  1. k-Means ClusteringConcept30 min
  2. Hierarchical ClusteringConcept25 min
  3. DBSCANConcept30 min
  4. Gaussian Mixture ModelsConcept35 min
  5. Principal Component AnalysisConcept30 min
  6. t-SNE & UMAPConcept30 min
  7. Anomaly DetectionConcept30 min

Module 12

Time series

Forecasting with trend and seasonality, and honest backtests.

  1. Time-Series Forecasting BasicsConcept30 min

Module 13

Evaluation metrics

Which number to trust, and when.

  1. Classification MetricsConcept30 min
  2. ROC-AUC vs PR-AUCConcept25 min
  3. Thresholds & Misclassification CostsConcept30 min
  4. Regression MetricsConcept30 min
  5. Ranking MetricsConcept25 min
  6. Probability MetricsConcept25 min
  7. Choosing the Right MetricConcept25 min

Module 14

Core concepts

Generalisation, evaluation and validation.

  1. The Bias–Variance Trade-offConcept30 min
  2. Cross-Validation & Data LeakageConcept30 min
  3. Class ImbalanceConcept30 min
  4. Hyperparameter TuningConcept30 min

Deep Learning

Module 15

Neural network basics

Neurons, MLPs, activations, losses and backpropagation.

  1. From Perceptron to MLPConcept20 min
  2. Activation FunctionsConcept20 min
  3. Loss Functions for Deep LearningConcept25 min
  4. Backpropagation Through a NetworkConcept25 min

Module 16

Training deep nets

Initialisation, vanishing gradients, normalisation, schedules and batch size.

  1. Weight InitialisationConcept20 min
  2. Vanishing & Exploding GradientsConcept25 min
  3. BatchNorm vs LayerNorm vs RMSNormConcept25 min
  4. Learning-Rate SchedulesConcept20 min
  5. Batch Size and Learning RateConcept20 min

Module 17

Regularisation and debugging

Dropout, weight decay, augmentation, early stopping and debugging training.

  1. DropoutConcept20 min
  2. Weight Decay vs L2, and AdamWConcept20 min
  3. Data Augmentation and MixupConcept20 min
  4. Early Stopping and Label SmoothingConcept20 min
  5. Debugging Training: Loss Won't Go DownInterview25 min
  6. Debugging Training: It Trains, But Something's OffInterview25 min

Module 18

CNNs

Convolutions, pooling, ResNets and transfer learning.

  1. Convolution and Receptive FieldConcept25 min
  2. Pooling, Stride, Padding and Output ShapesConcept20 min
  3. ResNets and Skip ConnectionsConcept20 min
  4. Transfer Learning and Fine-TuningConcept20 min

Module 19

Sequence models

RNNs, LSTMs, GRUs and seq2seq with attention.

  1. RNNs and Backprop Through TimeConcept20 min
  2. LSTM and GRUConcept25 min
  3. Seq2seq and Bahdanau AttentionConcept20 min

Module 20

Attention and Transformers

Self-attention, multi-head attention, positional encodings and the full Transformer.

  1. Self-Attention (Q, K, V)Concept25 min
  2. Multi-Head AttentionConcept25 min
  3. Positional Encodings (incl. RoPE)Concept25 min
  4. The Transformer BlockConcept20 min
  5. Encoder vs Decoder vs Encoder-DecoderConcept20 min
  6. The Transformer, End to EndConcept30 min

Module 21

Representation and generative models

Embeddings, VAEs, GANs, diffusion and contrastive learning.

  1. Embeddings and Representation LearningConcept20 min
  2. Autoencoders and VAEsConcept25 min
  3. GANsConcept25 min
  4. Diffusion ModelsConcept25 min
  5. Contrastive Learning (SimCLR, CLIP)Concept25 min

NLP Fundamentals

Module 22

Text basics

Preprocessing, bag-of-words, TF-IDF and text classification.

  1. Text PreprocessingConcept20 min
  2. Bag-of-Words, N-grams and TF-IDFConcept20 min
  3. Text Classification: Naive Bayes vs Logistic RegressionConcept25 min

Module 23

Language models and tokenisation

N-gram models, perplexity and subword tokenisers.

  1. N-gram Language Models and PerplexityConcept25 min
  2. Subword Tokenisation (BPE, WordPiece, SentencePiece)Concept25 min

Module 24

Tagging, retrieval and evaluation

Named entities, search and how to score generated text.

  1. Sequence Labelling and NERConcept20 min
  2. Search: BM25, Dense and Hybrid RetrievalConcept25 min
  3. NLP Evaluation: BLEU, ROUGE, EM/F1Concept20 min
  4. NLP Fundamentals PracticeInterview30 min

LLMs

Module 25

Modern architecture

What modern LLMs changed: RMSNorm, SwiGLU, GQA, MLA, FlashAttention, long context and MoE.

  1. Anatomy of a Modern LLMConcept20 min
  2. Attention Variants: MQA, GQA and MLAConcept25 min
  3. FlashAttentionConcept25 min
  4. Long Context: RoPE Scaling, Sliding Windows and Attention SinksConcept25 min
  5. Mixture of ExpertsConcept25 min

Module 26

Pretraining

Next-token prediction, data pipelines, scaling laws, distributed training and stability.

  1. The Next-Token ObjectiveConcept20 min
  2. Pretraining Data: Deduplication, Filtering and MixturesConcept25 min
  3. Scaling Laws: Kaplan vs ChinchillaConcept25 min
  4. Training at Scale: Mixed Precision, Parallelism, ZeRO/FSDPConcept25 min
  5. Training Instabilities: Loss Spikes, z-loss and QK-normConcept20 min

Module 27

Post-training

SFT, RLHF and reward models, DPO, reasoning models, distillation and LoRA.

  1. SFT and Instruction TuningConcept20 min
  2. Reward Models and RLHF (PPO)Concept25 min
  3. DPO and Preference OptimisationConcept25 min
  4. Reasoning Models: RLVR, GRPO and Test-Time ComputeConcept25 min
  5. Distillation: How Small Models Learn from Big OnesConcept20 min
  6. LoRA and QLoRA: Parameter-Efficient Fine-TuningConcept25 min

Module 28

Generation

Decoding and sampling, prompting and chain-of-thought, and hallucination.

  1. Decoding: Temperature, Top-k, Top-p and Min-pConcept20 min
  2. In-Context Learning and Chain-of-ThoughtConcept20 min
  3. Hallucination: Why LLMs Make Things UpConcept20 min

Module 29

Evaluating LLMs and practice

Benchmarks, contamination, LLM-as-a-judge and two practice sets.

  1. Benchmarks and ContaminationConcept20 min
  2. LLM-as-a-Judge and Human PreferenceConcept20 min
  3. LLM Practice 1: Pretraining and Scaling MathsInterview35 min
  4. LLM Practice 2: Post-training and GenerationInterview35 min

Next track

Next track: AI Engineering

How LLM applications are served, built and run, from inference to retrieval and agents.

Open AI Engineering