LLM fundamentals, RAG, evaluation, and applied AI engineering.
This is a spoken, conceptual discussion, not a live-coding exercise. You explain your reasoning out loud — concepts, tradeoffs, and "what would happen and why" — while the interviewer follows up and pushes on anything vague or surface-level.
Can you explain the difference between supervised, unsupervised, and reinforcement learning, with an example of each?
Can you explain what a token is in the context of an LLM, and why that matters for cost and context limits?
Can you explain what an embedding is, at a high level, and what it means for two embeddings to be 'close' to each other?
Can you explain the difference between prompting a model and fine-tuning it, and when you'd reach for each?
How would you go about evaluating whether an LLM-powered feature is actually working well, beyond just eyeballing outputs?
If you were adding an AI-powered feature to an existing product, how would you decide what should be handled by the model versus deterministic code?
Can you explain what a hallucination is in the context of an LLM, and why it happens?
Can you explain the difference between calling a hosted model API versus self-hosting a model, in terms of operational tradeoffs?
These are a few examples — each real session pulls 5 questions at random from a larger pool across every category above, so repeat sessions won't repeat the same set.
5 free interviews every month, any track, no card required.