Technical practice
Suggested practice: How would you design an agent that completes a multi-step task using tools?
What to coverThe role involves tool use, function calling, multi-step reasoning and orchestration. Explain how the agent would choose actions, track progress and stop or escalate when it cannot complete the task.
Suggested practice: How would you handle a failing or unreliable tool call in an agent workflow?
What to coverThe posting calls for tool integration and reliable production AI features. Discuss validation, timeouts, retries and safeguards against repeating an action that should run only once.
Suggested practice: How would you build an AI feature across the backend and front end?
What to coverThe role spans backend services, APIs and responsive interfaces. Describe the request flow, how the interface would handle a slow or failed response, and where you would test the integration.
Suggested practice: How would you design a retrieval-augmented generation pipeline?
What to coverThe posting names RAG, embeddings and vector databases. Explain document preparation, retrieval, use of retrieved context in a prompt, and how you would check whether the retrieved material supports the answer.
Suggested practice: How would you test a prompt pipeline whose outputs vary?
What to coverThe role calls for prompt pipelines and evaluation and testing of AI outputs. Suggest a representative test set, clear quality criteria and checks for regressions when prompts or models change.
Suggested practice: How would you integrate and compare model providers in a production feature?
What to coverThe posting mentions third-party APIs, OpenAI, Anthropic and open-source models. Discuss API boundaries, differences in capabilities and output formats, and how you would evaluate a provider change without assuming the models are interchangeable.
Suggested practice: How would you balance latency, cost, reliability and accuracy?
What to coverThe posting explicitly names all four as optimization goals. Walk through what you would measure, then explain a trade-off such as using a smaller model or fewer calls only if evaluation shows acceptable output quality.
Suggested practice: How would you divide application data between a conventional database and a vector store?
What to coverThe requirements mention SQL or NoSQL databases and familiarity with vector stores. Explain which data needs structured queries or application state, which needs similarity search, and how you would keep records and embeddings consistent.
Suggested practice: How would you ship and maintain an AI feature using cloud and CI/CD practices?
What to coverThe role requires comfort with a cloud platform and CI/CD, alongside testable, documented code. Describe deployment checks, configuration and secret handling, monitoring, and a rollback approach as preparation topics rather than a prescribed company process.
Suggested practice: When might fine-tuning be preferable to improving a prompt or RAG pipeline?
What to coverFine-tuning is listed as nice to have, while prompt pipelines and RAG are core duties. Compare the kinds of problems each approach could address, and discuss data, evaluation and maintenance costs before choosing one.
Behavioural practice
Suggested practice: Tell me about a feature you owned from concept to production.
What to coverThe posting emphasizes end-to-end feature ownership. Use a real situation, your specific actions and the result; explain how you decided the feature was ready to release.
Suggested practice: Tell me about a time you translated product or design needs into an implementation.
What to coverThe role involves collaboration with product and design teams. Describe the requirement, how you resolved uncertainty or competing priorities, what you built and the outcome.
Suggested practice: Tell me about a code review that improved the quality of a feature.
What to coverThe posting calls for code reviews and clean, testable, documented code. Give a real example of feedback you gave or received, the change you made and its result.
Suggested practice: Tell me about a difficult production issue you investigated.
What to coverThe role emphasizes problem-solving and the reliability of AI features. Describe the symptoms, how you narrowed down the cause, your action and the observed result without assuming a particular incident type.
Suggested practice: Tell me about a time you had to make an engineering trade-off under constraints.
What to coverThe posting calls for balancing latency, cost, reliability and accuracy. Use a real situation to explain the options, evidence behind your decision, action taken and outcome.