GenAI interview prep

Generative AI Interview Questions: LLMs Without the Hype

Generative AI interviews test how LLMs work (tokens, transformers, attention), how to adapt them (prompting, RAG, fine-tuning), and how to ship them safely (evaluation, hallucination control, cost and latency). Expect a system design question about an LLM-powered feature.

Practise with AI Interview Prep See AI engineer resume

16 questions answered
LLM fundamentals
RAG and fine-tuning
Production concerns

Model answers for 4 GenAI questions

Typical round: fundamentals, applied design, a past project deep dive

Q: When would you use RAG instead of fine-tuning? A: Use RAG when the model needs fresh or private knowledge that changes often, and you want citations. Use fine-tuning to change behaviour, style or format consistently, or to teach a narrow task. Many production systems combine both.

Q: What causes hallucinations and how do you reduce them? A: The model predicts likely text, not verified facts, so it can produce confident errors. Reduce them by grounding answers in retrieved sources, asking it to cite and abstain when unsure, constraining output formats, and evaluating with a labelled test set.

Q: What is a token and why does it matter? A: A token is a chunk of text, often part of a word, that the model processes. Context limits, latency and cost are all measured in tokens, so prompt length and retrieved context have to be budgeted.

Q: How do you evaluate an LLM feature? A: Build a test set of real inputs with expected outputs, score with automatic checks plus an LLM or human grader for quality, track regressions on every prompt or model change, and monitor production feedback.

How LLMs work
Tokens, embeddings, attention and next-token prediction explained simply show genuine understanding.
Adaptation choices
Prompting, RAG, fine-tuning and tool use: when each fits, and what each costs.
Production trade-offs
Latency, cost per request, context limits and caching separate demos from shipped features.
Evaluation and safety
Test sets, guardrails, prompt injection and data privacy are now standard interview topics.

Fundamentals

What is a large language model?
What is the transformer architecture?
What are embeddings?
What is temperature in sampling?
What is a context window?

Applied GenAI

RAG vs fine-tuning?
What is LoRA?
How do you reduce hallucinations?
What is prompt injection?
How do you choose a model size?

System design prompts

Design a customer-support assistant over company docs.
Design a resume-screening helper with fairness checks.
Cut the cost of an LLM feature by half.
Add evaluation to an existing chatbot.
Handle PII in prompts and logs.

Bring one project deep diveBe ready to explain the data, model choice, prompts, evaluation and what failed in one GenAI project you built.
Talk in trade-offsFor every choice, mention quality, latency, cost and risk. Interviewers listen for this balance.
Know the vocabulary preciselyTokens, embeddings, vector database, re-ranking, LoRA and RLHF should each have a crisp one-line definition.
Show evaluation habitsA test set and a regression check on prompt changes is one of the strongest signals you can give.
Rehearse aloudBrainUp AI Interview Prep runs AI-role mock interviews and gives feedback on clarity and depth.

What is generative AI?

Generative AI refers to models that create new content such as text, images, audio or code by learning patterns from large datasets. LLMs are the text-focused type.

What is a large language model?

An LLM is a neural network, usually a transformer, trained on large text corpora to predict the next token. With scale and fine-tuning it can follow instructions, summarise, write code and reason over text.

What is the transformer architecture?

A neural network design built on self-attention, which lets each token weigh every other token in the input. It processes sequences in parallel and underpins most modern LLMs.

What is attention in transformers?

Attention computes how much each token should focus on every other token, using query, key and value vectors, so the model captures context and long-range relationships.

What are embeddings?

Embeddings are numeric vectors that represent text or other data so that similar meanings sit close together. They power semantic search and retrieval in RAG.

What is a token?

A token is a unit of text the model reads and writes, often a word piece. Context limits, cost and speed are measured in tokens.

What does temperature control?

Temperature scales randomness in sampling. Low values make output more deterministic; higher values make it more varied and creative.

What is RAG?

Retrieval-augmented generation retrieves relevant documents for a query and adds them to the prompt, so the model answers from current or private data with citations.

What is fine-tuning?

Fine-tuning further trains a pre-trained model on task-specific examples to change its behaviour, format or domain performance.

What is LoRA?

Low-Rank Adaptation fine-tunes a model by training small added matrices instead of all weights, cutting memory and cost while keeping most of the quality.

What is RLHF?

Reinforcement learning from human feedback trains a reward model from human preference rankings, then optimises the LLM to produce responses people prefer.

What are hallucinations?

Hallucinations are fluent but false or unsupported outputs. Grounding with retrieval, citations, abstaining when unsure and evaluation help reduce them.

What is prompt injection?

Prompt injection is malicious text in user input or retrieved content that tries to override the system's instructions. Defences include input separation, least-privilege tools and output checks.

How do you evaluate LLM outputs?

Use a curated test set, automatic metrics or checks, LLM-as-judge or human review for quality, and track results on every model or prompt change.

How do you reduce LLM cost and latency?

Use smaller models where possible, shorten prompts, cache repeated results, stream responses, batch requests and route only hard cases to larger models.

How should I prepare for a generative AI interview?

Revise transformer basics, RAG, fine-tuning and evaluation, build one end-to-end LLM project with measured results, and practise a system design question for an LLM feature.

Practise GenAI interview answers with feedback

BrainUp AI Interview Prep asks AI-role questions and scores your answers.

Start interview practice Browse templates

The resume builder and PDF downloads are free. AI tools are part of BrainUp Premium.

Generative AI Interview Questions: LLMs Without the Hype

Explain trade-offs, not buzzwords.

Model answers for 4 GenAI questions

Built for scanners, humans and hiring intent.

How LLMs work

Adaptation choices

Production trade-offs

Evaluation and safety

Copy the structure, not the wording.

Fundamentals

Applied GenAI

System design prompts

Small edits that lift response rates.

Common questions, direct answers.

Practise GenAI interview answers with feedback

Generative AI Interview Questions: LLMs Without the Hype

Explain trade-offs, not buzzwords.

Model answers for 4 GenAI questions

Built for scanners, humans and hiring intent.

How LLMs work

Adaptation choices

Production trade-offs

Evaluation and safety

Copy the structure, not the wording.

Fundamentals

Applied GenAI

System design prompts

Small edits that lift response rates.

Common questions, direct answers.

Practise GenAI interview answers with feedback