Technical practice
How would you choose between a conventional ML model, a deep-learning model, and an LLM for a business use case?
What to coverSuggested answer focus: compare the task, available data, evaluation criteria, latency and implementation effort. Explain how you would establish a baseline before adopting a more complex approach.
How would you prepare data and engineer features before training a model?
What to coverSuggested answer focus: discuss data quality, missing values, leakage and features that reflect the task. Explain how you would keep training and evaluation data separate and check whether preprocessing improves results.
How would you evaluate and improve an ML model that performs poorly on new data?
What to coverSuggested answer focus: choose metrics suited to the use case and inspect errors rather than relying on one aggregate score. Discuss possible causes such as data quality or overfitting, then describe a controlled way to test improvements.
How would you design and test prompts for an LLM-based application?
What to coverSuggested answer focus: describe clear instructions, appropriate context and examples where useful. Explain how you would test outputs on varied inputs and account for unsupported answers, inconsistency and cost.
How would you build a RAG pipeline for answering questions from business documents?
What to coverSuggested answer focus: walk through document preparation, retrieval, context assembly and answer generation. Explain how you would evaluate both retrieval relevance and whether answers are supported by the retrieved material.
What trade-offs would you consider when selecting and using a vector database for RAG?
What to coverSuggested answer focus: discuss embedding and indexing choices, metadata filtering, update needs, retrieval quality and operational complexity. FAISS, ChromaDB and Pinecone are examples named in the posting; avoid claiming hands-on experience with any you have not used.
How would you design an AI agent to automate a business workflow safely?
What to coverSuggested answer focus: define the agent’s goal, available actions and points where a person should review its work. Discuss how you would handle tool failures, limit unintended actions and measure whether automation helps.
How would you decide whether to use LangChain, LangGraph, MCP or a simpler implementation?
What to coverSuggested answer focus: compare what the workflow needs—such as orchestration, state or tool connections—against added complexity. The posting names these ecosystems and Claude Code; explain only the tools you know, and identify what you would need to learn for the others.
How would you turn a Python AI prototype into a maintainable REST API?
What to coverSuggested answer focus: describe request validation, clear code structure, error handling and tests for both application and model-related behaviour. Explain how Git, Linux and SQL might support development or data access where relevant, without assuming a particular production setup.
A RAG application returns an irrelevant answer: how would you debug it?
What to coverSuggested answer focus: trace the input through retrieval, selected context, prompt and generated response to locate the failure. Compare fixes using representative test questions so that a prompt change does not mask a retrieval problem.
Behavioural practice
Tell me about a time you worked with people from another discipline to solve a technical problem.
What to coverSuggested answer focus: use a real situation, your specific actions and the result. Explain how you clarified the need, handled differing perspectives and checked whether the solution was useful.
Describe a time you took ownership of a task with an unclear starting point.
What to coverSuggested answer focus: set out the situation, how you defined the next steps and the result. Include how you surfaced uncertainties or blockers rather than waiting for detailed instructions.
Give an example of debugging a difficult issue under time pressure.
What to coverSuggested answer focus: describe the symptoms, the evidence you gathered, the tests you ran and the outcome. Explain how you prioritized a reliable fix over unverified guesses.
Tell me about a time you learned an unfamiliar technology to complete a practical task.
What to coverSuggested answer focus: explain the task, how you evaluated and learned the technology, what you built or tested, and the result. Distinguish what you verified firsthand from what you still needed to investigate.
Describe a time you changed your approach after feedback or evidence showed limited impact.
What to coverSuggested answer focus: give the situation, the feedback or measure that prompted the change, your actions and the result. Explain the trade-off you made and how you assessed whether the revised approach better served the goal.