Question bank, AI-assisted, AI system design
Design an AI support assistant
Design a support assistant that answers customer questions from our help center and past tickets, using an LLM. It must not make things up.
40 minutes, for mid-level, senior engineers
How the interview runs
- 1.Requirements. Before we design anything: what are the core functional requirements, and which non-functional ones matter most here?
- 2.Estimates. Let's size it. Assume 2 million questions a month, a help center of 5,000 articles and 3 million past tickets.
- 3.API + data model. What does the API look like, and how would you model and store the data?
- 4.High-level design. Walk me through the high-level architecture, request by request.
- 5.Deep dive. How do you make answers grounded and measurably good, and what happens when retrieval finds nothing useful?
- 6.Trade-offs + failure. What are the main trade-offs you made, and what happens when a component fails?
What you're graded on
- Clarified requirements15 pts
- Back-of-envelope estimates10 pts
- Clear API10 pts
- Data model + storage choice15 pts
- Coherent architecture15 pts
- Grounding, evals + fallbacks20 pts
- Trade-offs8 pts
- Failure handling7 pts
Practice this one for real
Answer it phase by phase, then get the rubric score, what you missed and the model answer outline.
Unlock with Lifetime