Interview
Prepare with real answers
76 questions across 10 categories. Every entry is a draft until it passes review; provenance is shown on each page.
Categories
- LLM & transformer fundamentals9
- Prompting & context engineering4
- Embeddings & retrieval5
- RAG & retrieval pipelines21
- Agents, tools & orchestration0
- Agentic coding0
- MCP, A2A & interoperability0
- Fine-tuning & adaptation5
- Evaluation & quality0
- Security & governance0
- Production & LLMOps0
- System design0
- Multimodal, vision & voice5
- Data & storage4
- Maths & CS fundamentals14
- Frameworks & tooling0
- Take-homes & career3
- Behavioural & project deep-dives6
- Practical coding0
All questions
- Walk me through an AI feature you shipped end to end and your exact role. What decisions were yours?behavioral · mid
- Tell me about a time a model surprised you in production. How did you find out, and what did you change?behavioral · mid
- Describe a time you disagreed with a product or engineering stakeholder about using an agent, fine-tuning, or RAG.behavioral · senior
- Tell me about a model-quality regression you introduced or discovered after deployment. What did you do?behavioral · senior
- How do you decide when a model or agent is good enough to ship?behavioral · staff
- Tell me about a time you chose a simpler model or architecture over the more sophisticated option.behavioral · senior
- Take-home: build a RAG assistant over a document set with citations. What do you build first, and how do you scope it?career · mid
- Defending your take-home: if you had to simplify the system, which component would you remove, and what would you lose?career · senior
- Take-home: implement a minimal workflow engine with graph nodes, state, branching, looping and a 50-step cap. What does a strong submission contain?career · mid
- How do you choose a chunk size, and what concretely goes wrong at each extreme?data · junior
- Users say a rate-limits answer quotes correct numbers but never says what they apply to. You trace it to one chunk. Diagnose and fix.data · senior
- How do you ingest a 200-page PDF of charts and tables where the answer exists only inside a line chart, with no matching text?data · senior
- Page 1 of a financial report says "all amounts in thousands". How do you stop chunks from losing that document-wide context?data · senior
- Where do embeddings systematically fail as a retrieval mechanism, and which failures can you fix cheaply?embeddings · mid
- You must choose an embedding model for a 900k-chunk support corpus. Walk me through the bake-off and the constraints that don't appear on a leaderboard.embeddings · senior
- You have 25M chunks at 1,536 dimensions. Size the index and tell me how you make it fit — then what changes at 10x.embeddings · staff
- Why does swapping the embedding model force a full re-index, and how do you roll that out without downtime?embeddings · senior
- Cosine similarity versus dot product: why normalise, and what breaks if you don't?embeddings · junior
- The PM says: fine-tune the model on our docs so it knows our product. How do you respond?fine-tuning · junior
- Full fine-tuning or LoRA? Walk me through how you decide, including the memory numbers.fine-tuning · mid
- Explain LoRA mathematically. What do r and alpha control, and which modules would you target?fine-tuning · mid
- DPO versus PPO-based RLHF: when would you choose each, and what does DPO give up?fine-tuning · senior
- After a DPO run, reviewers say answers got longer and more flattering but not more accurate. What happened, and what do you do?fine-tuning · senior
- Why did transformers replace RNNs and LSTMs for language modeling?fundamentals · junior
- Walk me through what happens inside a single transformer decoder block.fundamentals · junior
- Explain self-attention step by step. What exactly are Q, K, and V, and why are K and V separate?fundamentals · junior
- Why does scaled dot-product attention divide by √d_k?fundamentals · mid
- What is causal masking, why is it needed, and why apply it before the softmax?fundamentals · junior
- Why use multiple attention heads instead of one head of the full model width?fundamentals · mid
- Pre-norm versus post-norm: what changes, and why did essentially everyone move to pre-norm and RMSNorm?fundamentals · senior
- What does the feed-forward network do, where do most parameters live, and why is SwiGLU the default?fundamentals · mid
- Why is RoPE called a relative position scheme even though each token is rotated by its absolute index, and how do you extend its context?fundamentals · senior
- What is FlashAttention, why is it exact rather than approximate, and what does it not fix?fundamentals · senior
- Walk me through BPE training and encoding. Why start from bytes, and what does vocabulary size trade off?fundamentals · mid
- Why do LLMs use subword tokenization rather than words or characters, and what does that choice cost you?fundamentals · junior
- If your vocabulary is 100K tokens, how do you reason about whether that is the right size, and what does it cost?fundamentals · senior
- What does perplexity measure, and why can't you compare perplexity across models with different tokenizers?fundamentals · senior
- What is grouped-query attention, how does it differ from multi-head and multi-query attention, and what does it buy you?llm-fundamentals · senior
- Explain the KV cache. What does it store, why does it speed up generation, and how would you estimate its memory?llm-fundamentals · mid
- Explain prefill versus decode in LLM inference. Which phase is compute-bound and which is memory-bandwidth-bound, and why does that matter for serving?llm-fundamentals · mid
- Explain speculative decoding. Why is it lossless, and when does it stop helping?llm-fundamentals · senior
- Your GPU has free KV-cache memory but throughput is poor and requests sit idle waiting for a batch to finish. What is the lever, and how does it differ from PagedAttention?llm-fundamentals · staff
- What does temperature actually do, give the formula, and why is temperature 0 not a determinism guarantee?llm-fundamentals · junior
- Compare greedy, top-k, top-p and min-p sampling. Why truncate the distribution at all?llm-fundamentals · mid
- How can you increase the context length of an LLM, and what actually limits it?llm-fundamentals · staff
- Your transformer's KV cache grows too large during long sequence generation and you run out of memory. Walk me through diagnosis and mitigation.llm-fundamentals · senior
- Argue both sides: cascaded ASR plus LLM plus TTS versus a native speech-to-speech model for a production voice agent.multimodal · senior
- Budget the latency for a real-time voice agent: VAD, ASR, LLM, TTS, network. Where does the time go, and what do you optimise first?multimodal · senior
- A video summariser samples one frame every three seconds. A customer summary says an export may not have finished, but the recording shows an Export ready toast. What went wrong, and how do you fix it?multimodal · mid
- When would you use a multimodal model instead of OCR plus an extraction pipeline, and what are its failure modes?multimodal · junior
- How do you evaluate ASR quality beyond word error rate, and TTS quality when there is no single correct output?multimodal · senior
- You stuff 60 retrieved documents into one prompt and the model uses facts from the beginning and end but ignores a crucial fact in the middle. Diagnose it and fix it.prompting · mid
- An agent reads retrieved documents and tool outputs and can take actions. How do you defend against indirect prompt injection, and why can't a better system prompt solve it?prompting · staff
- Your JSON parsing works in development but fails intermittently in production even at temperature 0. Diagnose it and describe the durable fix.prompting · mid
- Your chain-of-thought prompt is not improving accuracy on a reasoning task. What do you check and what do you change?prompting · mid
- Explain parent-child (small-to-big) retrieval. What problem does it solve and what does it cost you?rag · mid
- Explain HNSW. What do M, ef_construction and ef_search each control, and which can you change in production?rag · mid
- How would you tune HNSW for 10 million vectors when the product needs p95 retrieval under 100 ms?rag · staff
- IVF, HNSW, product quantisation — pick an index for 500M vectors under a hard memory budget and defend it.rag · senior
- Each user can see under 1% of a 50M-document corpus. What happens if you apply the permission filter after ANN search, and what do you do instead?rag · staff
- Pasting an order ID like ORD-88213 returns generic refund-policy chunks after a move from keyword to vector search. Diagnose and fix.rag · mid
- Weighted min-max fusion with an alpha versus reciprocal rank fusion — which do you ship, and how do you tune it?rag · senior
- The vector database already returns similarity scores. Why add a reranker, and how deep do you rerank?rag · mid
- When does reranking not help, or actively hurt, a production RAG system?rag · senior
- How do you evaluate retrieval separately from generation, and which metric do you fix first?rag · mid
- Your offline recall is 0.91 but production recall is 0.40. What went wrong in the evaluation, and how do you fix it?rag · staff
- You want an LLM judge to gate deploys on faithfulness. How do you validate it before you trust it?rag · senior
- Your recall@5 is 95%. Does that mean the system answers 95% of questions correctly?rag · senior
- Latency, throughput and retrieval health all look normal, but users say answers are subtly out of date. Diagnose.rag · staff
- A junior analyst gets an answer that quotes a privileged memo they are not staffed on. Firewalls and DLP saw nothing. What failed?rag · staff
- Answers cite real documents that do not actually contain the claim. How do you detect and fix bad citations?rag · senior
- After adding query rewriting and a reranker, p95 went from 900 ms to 4 s. Where do you look and what do you cut?rag · senior
- Walk me through a production RAG pipeline end to end. Where do these systems actually fail?rag · senior
- A question needs facts from three documents, like "compare UK and Netherlands parental leave". How does retrieval change?rag · senior
- What does the retriever do when a query has no relevant documents at all? Design the abstain path.rag · senior
- Your RAG system works on 10k documents but degrades at 1M. What changes besides the index?rag · staff