Technical practice
How would you choose and evaluate an ML approach for a business problem?
What to coverSuggested answer focus: Define the objective, establish a baseline, and choose evaluation measures that reflect the intended use. Explain how you would detect overfitting and decide whether the result is useful enough to pursue.
How would you improve an LLM application through prompt engineering?
What to coverSuggested answer focus: Describe how you would structure instructions and examples, then test changes against representative inputs. Discuss how you would assess answer quality, consistency, and failure cases when working with models such as OpenAI or Claude.
When would you consider fine-tuning or multimodal AI instead of a prompt-only approach?
What to coverSuggested answer focus: Compare the problem each approach could address and the additional data, evaluation, and maintenance it may require. Fine-tuning and multimodal AI are listed in the posting; a practical decision framework is a suggested preparation topic, not a stated employer preference.
How would you design and troubleshoot a RAG solution?
What to coverSuggested answer focus: Walk through document preparation, retrieval from a vector database, and using retrieved context in a response. Explain how you would investigate irrelevant retrieval or answers unsupported by the retrieved material.
How would you control tool use and memory in an agentic workflow?
What to coverSuggested answer focus: Describe when function calling, memory, or multiple agents would help, and when a simpler workflow would suffice. Discuss how you would handle failed tool calls, limit unintended actions, and test the workflow; familiarity with frameworks such as LangChain, Strands, or CrewAI is stated in the posting.
How would you connect a Python AI service to a ReactJS and NodeJS application?
What to coverSuggested answer focus: Outline the responsibilities of each component and how requests, responses, and errors would move between them. Discuss trade-offs in keeping model-facing logic separate from the user interface and testing the integration.
How would you assess whether an AI proof of concept is ready to advance?
What to coverSuggested answer focus: Define test cases and success measures tied to the business problem, including examples where the solution fails. Explain what evidence you would share with data, engineering, and business partners before investing in production work.
What would you put in place to deploy and monitor an AI solution?
What to coverSuggested answer focus: Discuss CI/CD, repeatable testing, deployment checks, and monitoring of application and model behaviour. Relate GitHub and Jenkins to that workflow without assuming the posting specifies a particular pipeline design.
How would you evaluate governance and bias risks in a proposed AI use case?
What to coverSuggested answer focus: Identify who could be affected, what harmful or uneven outcomes to test for, and how you would document findings. Explain how evaluation and ongoing monitoring could inform mitigation rather than treating a successful prototype as sufficient.
How would you compare AI service options for a prototype moving toward production?
What to coverSuggested answer focus: Consider model capabilities, integration effort, evaluation results, and operational needs when comparing options. AWS Bedrock experience and exposure to Azure or GCP AI services are preferred, so platform-specific comparisons are a suggested preparation topic, not a stated requirement for every solution.
Behavioural practice
Tell me about a time you defined an AI or analytics problem with business and technical partners.
What to coverSuggested answer focus: Give the situation, your actions to clarify the problem and align the teams, and the result. Explain how you handled differing priorities or definitions of success.
Describe a time an experiment or proof of concept did not meet its goal.
What to coverSuggested answer focus: Set out the situation, how you investigated the result and decided what to change or stop, and what happened next. Use a real example rather than implying that every experiment led to deployment.
Tell me about a time you helped move a technical solution toward production.
What to coverSuggested answer focus: Describe the starting state, your specific work with engineering or data partners on testing and operational readiness, and the result. If the work stopped before deployment, be clear about what was completed and learned.
Give an example of explaining a complex technical concept to a non-technical stakeholder.
What to coverSuggested answer focus: Describe the audience and decision at stake, how you adapted your explanation, and the outcome. Show how you communicated limitations as well as potential value.
Describe a time you raised a risk in an AI or data-driven project.
What to coverSuggested answer focus: Explain the situation, the risk you identified, how you discussed it with others, and the result. Responsible AI, governance, and bias mitigation appear in the posting; using a real example of those issues is a suggested preparation approach if applicable.