Technical practice
How would you design and evaluate an LLM-based application before deployment?
What to coverSuggested answer focus: Describe how you would define the task, choose an LLM-based approach and test output quality against representative examples. Discuss trade-offs such as response quality, latency and cost without assuming a particular model or deployment platform.
How would you build a scalable retrieval-augmented generation pipeline?
What to coverSuggested answer focus: Walk through document preparation, indexing, retrieval, context assembly and generation. Explain how you would measure answer quality and diagnose whether a poor answer came from retrieval or generation.
How would you choose and tune a vector-search component for RAG?
What to coverSuggested answer focus: Compare retrieval quality, update needs, latency and operational complexity, then explain how you would test changes. Vector databases, including Pinecone, FAISS and Weaviate, are listed as good-to-have experience rather than required choices.
How would you structure an agentic workflow using LangGraph?
What to coverSuggested answer focus: Explain states, transitions and when an agent should stop or hand off work. Discuss how you would make failures observable and decide whether an agentic workflow is preferable to a simpler fixed sequence.
How would you develop and assess an NLP solution for a real-world task?
What to coverSuggested answer focus: Describe how you would establish a baseline, select evaluation examples and inspect error patterns. Distinguish what conventional NLP methods and LLM-based methods might offer for the chosen task.
Where could knowledge graphs and semantic layers fit in an AI application?
What to coverSuggested answer focus: Explain how structured relationships and consistent meanings could support retrieval or downstream answers. Compare their potential value and maintenance cost with using unstructured text alone.
How would you expose an AI workflow through a Python and FastAPI API?
What to coverSuggested answer focus: Describe request validation, response design, error handling and how to keep slow AI work from degrading the API. Explain how you would test the boundary between the endpoint and the underlying workflow.
How would you investigate poor SQL or NoSQL database performance?
What to coverSuggested answer focus: Start with the access pattern and measurements, then discuss indexing, query design and data-model trade-offs. Explain how you would verify that an optimization helps without compromising correctness.
How would you design a scalable system around RAG services, agents and APIs?
What to coverSuggested answer focus: Describe service boundaries, data flow and likely scaling bottlenecks in an architecture using the components named in the posting. Discuss when separating microservices helps and when it adds unnecessary complexity.
How would you make changes to an AI application safe to release?
What to coverSuggested answer focus: Cover Git-based review, automated checks and CI/CD for code and AI-workflow changes. As a suggested extension, discuss deployment monitoring and rollback; MLOps and cloud-platform knowledge are listed as good to have.
Behavioural practice
Tell me about a time code review improved an AI or API change.
What to coverSuggested practice: Use a real situation, your actions in giving or receiving feedback, and the result. Focus on a specific technical concern and how the review affected the implementation.
Tell me about a time you diagnosed a poor result from an NLP or retrieval system.
What to coverSuggested practice: Set out the symptom and context, how you isolated the cause, and the result of your changes. If your example was not a RAG system, explain the transferable troubleshooting approach rather than presenting it as RAG experience.
Tell me about a time you weighed competing system-design options.
What to coverSuggested practice: Describe the situation, the options and trade-offs you evaluated, the action you took, and the observed result. An example involving scalability, databases or service boundaries would connect to the stated responsibilities.
Tell me about a time you improved a delivery or deployment practice.
What to coverSuggested practice: Give the starting situation, your contribution to a Git, CI/CD or release workflow, and the result. Be precise about what you personally changed and how you assessed its effect.
Tell me about a time you coordinated an API change with another system component.
What to coverSuggested practice: Use a real situation to explain the interface or integration issue, how you worked with others to resolve it, and the result. This is a suggested way to prepare for the posting's API and microservices work, not a stated interview topic.