Everything worth knowing

The probability, linear algebra and calculus that every ML interview quietly assumes you know.

Linear models, trees, ensembles and clustering. Still most of what runs in production.

Backprop, architectures, and everything that makes training actually work.

Tokenization through attention. The architecture every modern interview comes back to.

Pretraining, alignment, RAG and agents. The fastest moving part of any interview loop.

Detection, segmentation and the vision questions that follow the CNN warm up.

MDPs through to PPO. Asked more and more now that RL sits behind LLM alignment.

Retrieval, ranking and personalization. The best paid specialism in applied ML.

The 45 minute whiteboard round. Framing, data, serving, feedback loops, tradeoffs.

What happens after the notebook. Pipelines, monitoring, drift and rollbacks.

Precision, recall, AUC, and the question about which metric you would actually optimize.

Where most of your model quality comes from, and where interviewers dig for rigour.

NumPy, PyTorch, and the round where you implement it from scratch.

Fairness, privacy, and the non technical half of the loop that decides your level.