LLMs & AI interview questions

All interview topics

Interview questions on building with large language models: how they work, prompting, retrieval, agents, evaluation and safety.

50 questions
16 easy
28 medium
6 hard
Free account required

Every question is listed below. Sign in, or create an account in a few seconds, to open the full answers with code, diagrams and follow-ups.

50 of 50 questions

LLMs & AI interview questions

1What is a large language model, and how does it generate text?EasyFundamentals

2What are tokens, and why do they matter in practice?EasyFundamentals

3Explain the transformer architecture and self-attention at a high level.MediumFundamentals

4What is a context window, and what happens when a conversation exceeds it?EasyFundamentals

5What do temperature, top-p and top-k control?EasyFundamentals

6How are chat LLMs trained? Explain pre-training, instruction tuning and RLHF.MediumFundamentalsFine-tuning

7What are hallucinations, why do they happen, and how do you reduce them?EasyFundamentalsSafety

8What are embeddings, and what are they used for?EasyEmbeddingsRAG

9How do you choose which LLM to use for a feature?MediumProduction

10What are the trade-offs between open-weight and proprietary (API) models?MediumProduction

11Explain zero-shot, few-shot and chain-of-thought prompting.EasyPrompting

12What is a system prompt, and what should it contain?EasyPrompting

13How do you get reliable structured output (such as JSON) from an LLM?MediumPromptingProduction

14What is prompt injection, and how do you defend against it?MediumSafetyAgents

15What are reasoning (thinking) models, and when should you use them?MediumFundamentalsProduction

16What is retrieval-augmented generation (RAG), and why use it?EasyRAG

17Walk through the components of a production RAG pipeline.MediumRAG

18How do you choose a chunking strategy for RAG?MediumRAG

19What is a vector database, and how does approximate nearest neighbour search work?MediumRAGEmbeddings

20What is cosine similarity, and why is it used for embeddings?EasyEmbeddings

21What is hybrid search, and why add a re-ranker?MediumRAG

22How do you evaluate a RAG system?HardRAGEvaluation

23What are common RAG failure modes, and how do you fix them?HardRAG

24When would you fine-tune a model instead of using prompting or RAG?MediumFine-tuningRAG

25What is LoRA (parameter-efficient fine-tuning)?MediumFine-tuning

26What are quantisation and distillation?MediumFine-tuningProduction

27What is an AI agent, and how is it different from a workflow?EasyAgents

28How does tool (function) calling work?MediumAgents

29How do you design a reliable agent loop?HardAgentsProduction

30What is the Model Context Protocol (MCP)?MediumAgents

31When would you use a multi-agent system, and what are the trade-offs?HardAgents

32How do LLM applications handle memory across long conversations or sessions?MediumAgents

33How do you evaluate an LLM application?MediumEvaluation

34What is LLM-as-a-judge, and what are its pitfalls?MediumEvaluation

35What are the limitations of public LLM benchmarks?MediumEvaluation

36How would you reduce the cost and latency of an LLM feature?MediumProduction

37What is prompt caching, and when does it help?MediumProduction

38Why stream LLM responses, and what changes in your application?EasyProduction

39How do you handle rate limits, timeouts and errors from LLM APIs?MediumProduction

40How do you monitor an LLM application in production?MediumProductionEvaluation

41LLM outputs aren't deterministic. How do you test LLM-powered code?MediumEvaluationProduction

42What are guardrails in LLM applications?MediumSafetyProduction

43How do you handle sensitive and personal data when using LLM APIs?MediumSafetyProduction

44How can bias show up in LLM applications, and how do you mitigate it?MediumSafety

45What's the difference between supervised, unsupervised and reinforcement learning?EasyML basics

46What is overfitting, and how do you prevent it?EasyML basics

47Explain precision, recall and F1. When does each matter most?EasyML basicsEvaluation

48Design a customer support chatbot that answers questions from company documentation.HardSystem designRAG

49Design a system that uses an LLM to classify 1 million support tickets a day.HardSystem designProduction

50How would you explain how ChatGPT works to a non-technical stakeholder?EasyFundamentalsCommunication