All 24 lessons in curriculum order, grouped by track - from math foundations through LLMs, RAG and MLOps. Every lesson pairs five deep-dive questions with a quiz and a hands-on challenge.
Every embedding, attention head and gradient update is linear algebra. Master the five questions interviewers actually ask about vectors, matrix decompositions and derivatives.
18 min · 5 questions · foundational
The probability questions that separate candidates: base rates, why MLE is just cross-entropy in disguise, MAP as regularisation, and the distribution choices baked into every loss function.
17 min · 5 questions · foundational
Shipping a model means running an experiment. Learn how to size it, read it honestly, avoid peeking, and explain why offline AUC gains vanish online.
19 min · 5 questions · intermediate
The five questions that open almost every ML interview. Answer them with diagnostics and trade-offs instead of definitions and you clear the screen in ten minutes.
18 min · 5 questions · foundational
Interviewers still open with linear models because they expose whether you understand optimisation, probability and diagnostics - or just call .fit().
19 min · 5 questions · intermediate
Gradient boosting still wins most tabular problems in production. Know how it works internally, why it beats random forests, and how to explain its predictions.
20 min · 5 questions · intermediate
Picking the metric is a product decision disguised as a technical one. Learn to derive the right metric from the cost of each error type - and to defend it.
18 min · 5 questions · intermediate
Features still beat models on tabular problems. Learn the engineering patterns that win, the honest way to handle imbalance, and how to evaluate unsupervised models with no labels.
19 min · 5 questions · intermediate
Derive backprop, explain why ReLU replaced sigmoid, and know exactly which initialisation to use and why - the three things every deep-learning interview verifies.
20 min · 5 questions · intermediate
The practical half of deep learning: how to pick an optimiser, set a learning rate, stop a run from diverging, and regularise without destroying capacity.
19 min · 5 questions · advanced
The most-asked architecture question in AI interviews. Derive scaled dot-product attention, explain the √d, and know why grouped-query attention exists.
22 min · 5 questions · advanced
Everything downstream of the tokenizer depends on it. Understand BPE, the pre-training objectives, and the scaling laws that decide how big a model should be.
19 min · 5 questions · intermediate
When to fine-tune at all, how LoRA actually works, and the real difference between RLHF and DPO - with the numbers that justify each choice.
20 min · 5 questions · advanced
The half of LLM engineering that decides whether your product is profitable: decoding strategies, quantisation trade-offs, and how to get 10× throughput from the same GPU.
20 min · 5 questions · advanced
RAG is the most-built AI system in industry and the most-failed. Learn the design decisions that actually determine whether it works: chunking, hybrid retrieval and index choice.
20 min · 5 questions · advanced
Agents are easy to demo and hard to ship. Learn the loop, the failure modes, the guardrails, and when a deterministic pipeline beats an agent outright.
20 min · 5 questions · expert
The gap between a notebook and a service. Serving patterns, the feature-store problem they solve, and how to ship a model without breaking production.
19 min · 5 questions · advanced
Models decay silently. Learn what to monitor when labels arrive months late, how to distinguish drift types, and how to run an ML incident.
18 min · 5 questions · advanced
A repeatable structure for the 45-minute ML system design round, plus the three moves that separate a hire from a no-hire in the first five minutes.
20 min · 5 questions · expert
Four more worked designs covering the surfaces most companies actually build: search, feeds, ads and moderation - each with the trap the interviewer is waiting for.
20 min · 5 questions · expert
The ML coding round is not LeetCode. It is 'implement k-means in 20 minutes' and 'vectorise this loop' - with an interviewer watching how you debug.
20 min · 5 questions · intermediate
Write a correct training loop from memory, explain autograd precisely, and debug a broken model live - the three things a PyTorch round actually tests.
18 min · 5 questions · advanced
The round that decides your level. How to tell a project story that survives ten follow-up questions, and how to answer 'tell me about a failure' without hurting yourself.
17 min · 5 questions · intermediate
The meta-game: which companies test what, how to win a take-home, what to ask your interviewers, and how to negotiate without risking the offer.
18 min · 5 questions · intermediate