Section 6.4 · 71 questions

Interview question bank

Answer out loud first, then compare with what a good answer covers. Mark each one practising or confident – your mock-interview day is Saturday of Week 21.

Technical0/30

System design0/15

FDE0/10

Behavioural0/10

AI tools0/6

Technical

0 of 30 confident

Practice · question 1

List vs tuple vs dict vs set – when do you use each?

Or write your answer and have it marked

How these employers interview

LeetCode-style problems from Week 5 onwards and a practical, multi-level build (Week 21); a project you can defend in a deep-dive (P2/P5, Week 21); LLM system design (Week 21); and, for FDE, timed decomposition of a vague customer problem (Weeks 20–21).

Anthropic – SWE / AI-engineering roles

S12, S13, S14

Recruiter call → timed CodeSignal project (commonly ~90 min; the requirements build level by level) → live practical coding (~1 h) → onsite with practical coding, system design, a project deep-dive and a culture interview. The coding is practical rather than puzzle-style.

OpenAI – engineering / FDE

S15, S16, S17

Official guide: assessments vary by role (pair coding, take-home, HackerRank). The final round is 4–6 hours with 4–6 people over 1–2 days. Candidates single out a past-project walkthrough round.

Palantir – Forward Deployed Software Engineer

S18, S20, S23

A 60-min decomposition round (break a vague real-world problem into parts, with little or no code); a 60-min CodePair coding round with ~20 min of behavioural questions; a learning round. FDE loops may include a ‘Build to Apply’ take-home.

FDE roles generally

S19, S23

The decomposition / open-ended customer case is now used beyond Palantir (Aced, formerly Exponent, names OpenAI and Anthropic). It typically takes 45–90 minutes.

Mistral AI – Applied AI Engineer (London)

S21

One candidate’s report: HR call → LeetCode-medium coding round → interview with a team lead / senior AI engineer. Low confidence (a single report).

Google DeepMind – Research Engineer

S22

The research-engineer track is more flexible than the research-scientist track and accepts strong engineers with ML experience without a PhD.

Pay ranges, employers hiring and events are on the UK market page.

Technical

1List vs tuple vs dict vs set – when do you use each?

2Why is x in my_list slower than x in my_set ?

3Write a function that retries a failing API call.

4What is a virtual environment and a lockfile?

5Explain HTTP methods and status codes 200, 201, 400, 401, 404, 429, 500.

6Design REST endpoints for a ‘documents’ resource.

7SQL: WHERE vs HAVING; INNER vs LEFT JOIN. Write ‘top 5 boards by patients’.

8How do you test code that calls an external API?

9What is a merge conflict and how do you resolve it?

10What is a token and why does it matter?

11What does temperature do? When do you set it low?

12How do you get reliable JSON from an LLM?

13Walk through tool use / function calling end to end.

14What are embeddings and cosine similarity?

15Walk through a RAG pipeline and where it fails.

16How do you choose chunk size and overlap?

17Keyword (BM25) vs vector search – when hybrid?

18How do you evaluate a RAG system?

19What is LLM-as-judge and how do you trust it?

20What is prompt injection? Mitigations in RAG and agents?

21Agent vs workflow – how do you choose?

22How would you cut LLM cost and latency?

23What is a context window and what goes wrong with very long inputs?

24Explain gradient descent and the learning rate.

25Explain backpropagation simply.

26How do you detect and prevent overfitting?

27What does attention do in a transformer?

28What is LoRA and when would you fine-tune instead of using RAG or prompting?

29How is training on Apple Silicon different from NVIDIA GPUs?

30Docker image vs container – and how do you handle secrets?

How these employers interview