Design an Agentic Question-Answering System Over Personal Notes
Design an agentic system that answers users' questions based on their personal notes, using only two fixed tools (Get Note and Search Note) to retrieve and synthesize relevant information into accurate, grounded responses.
Question Timeline
See when this question was last asked and where, including any notes left by other candidates.
Early October, 2026
You’re given an agentic system with access to two tools, Get Note and Search Note, and you're asked to design a system that answers people’s questions based on their notes. The one hard constraint is that you can’t create any new tools, so the whole design has to work within those two. How to prepare. Since the tools are fixed, the interview is really about how well you reason about the agent itself. It’s worth thinking through: Tool use: how the agent decides when to search versus fetch a specific note, and how it plans multiple steps for a complex question. Retrieval: how it writes and refines search queries, handles vague questions, and combines information across several notes. Edge cases: what happens when search returns nothing relevant, too many results, or conflicting notes, and how the agent avoids making things up. Context limits: how to manage long notes or many results without overflowing the model’s context. Quality and evals: how you’d measure whether answers are correct and grounded in the notes, and how you’d improve the system over time. Practical trade-offs: latency, cost and reliability, given Chief’s product serves busy senior leaders.
Your account is free and you can post anonymously if you choose.