Technical practice
How would you design document ingestion and chunking for a RAG pipeline?
What to coverPosting: The role covers document ingestion and chunking in end-to-end RAG pipelines. Suggested preparation (inference): Explain how you would preserve useful context, handle document updates, and test whether chunk size and overlap improve retrieval.
How would you choose an embedding model and vector-indexing approach?
What to coverPosting: The role calls for hands-on work with embedding models and vector databases such as Chroma, Qdrant, Pinecone, or pgvector. Suggested preparation (inference): Discuss retrieval quality, indexing and update needs, latency, and operational complexity without assuming the company uses a particular database.
When would you add reranking to retrieval, and how would you evaluate it?
What to coverPosting: Retrieval, reranking, and measuring retrieval quality are stated responsibilities. Suggested preparation (inference): Describe a baseline, a relevant test set, and the quality-versus-latency trade-off you would check before keeping a reranker.
How would you structure a Python API service around an LLM-backed feature?
What to coverPosting: The role specifies Python, REST APIs, asynchronous programming, clean object-oriented code, and FastAPI or Python backend services. Suggested preparation (inference): Explain request validation, separation of concerns, async handling of external calls, and testing of failure paths.
How would you integrate and compare LLM API or open-source model options?
What to coverPosting: The role names Anthropic Claude, OpenAI, Gemini, and open-source models as integration options. Suggested preparation (inference): Explain how you would isolate provider-specific behavior and compare output quality, latency, cost, and error handling for a defined use case.
How would you make structured outputs and tool calls reliable?
What to coverPosting: Designing structured outputs, function or tool calling, and agent workflows is part of the work. Suggested preparation (inference): Cover schema validation, safe handling of tool inputs and failures, and when a simpler fixed workflow is preferable to an agent.
How would you evaluate a RAG answer and reduce hallucinations?
What to coverPosting: The role includes basic evaluations and guardrails to reduce hallucinations. Suggested preparation (inference): Describe test cases for answer support from retrieved material, how you would inspect failures, and what the system should do when evidence is insufficient.
What would you log to troubleshoot quality and API cost in an LLM application?
What to coverPosting: Logging, retrieval quality, and API cost monitoring are explicit responsibilities. Suggested preparation (inference): Identify useful per-request signals, such as retrieval results, errors, latency, and usage, while explaining how you would avoid logging sensitive content unnecessarily.
How would you use AI-assisted coding tools without weakening code quality?
What to coverPosting: The role calls for daily use of tools such as Claude Code, Cursor, or Copilot to explore codebases, write unit tests, and speed delivery, alongside rigorous inspection of generated code. Suggested preparation (inference): Walk through task scoping, reviewing changes, running tests, and using Git to keep changes traceable.
How would you prepare an LLM backend for deployment?
What to coverPosting: Shipping production applications is central; Docker, Linux, and basic AWS or GCP deployment are listed as good to have. Suggested preparation (inference): Discuss configuration, health checks, and a deployment smoke test at a basic level, without treating a specific platform as required.
Behavioural practice
Tell me about a time you shipped a working AI project beyond a basic chat application.
What to coverPosting: A shipped or working project such as semantic search, document RAG, or an automated agent is listed as proof of work. Suggested preparation (inference): Use a real example to explain the situation, your actions, and the result, including what made the project usable.
Tell me about a time you caught a problem in AI-generated code.
What to coverPosting: The role emphasizes guiding AI coding agents and rigorously inspecting generated code. Suggested preparation (inference): Give the situation, how you verified and corrected the code, and the resulting effect on correctness or delivery.
Tell me about a time you investigated poor results in a search or RAG system.
What to coverPosting: The work includes maintaining RAG pipelines and measuring retrieval quality. Suggested preparation (inference): If you have a relevant real example, describe the situation, diagnostic actions, and measured result; otherwise, do not invent an incident.
Tell me about a time you balanced delivery speed with software reliability.
What to coverPosting: The role emphasizes building, testing, and shipping clean production software quickly. Suggested preparation (inference): Choose a real situation, explain the trade-off and safeguards you chose, and state the result without implying the employer follows a particular release process.