Design a Law Firm Memo Q&A AI Agent

Problem Overview

Design an AI agent that answers questions based on big law firms' memos. Large firms continuously publish client memos (also called client alerts or legal updates) on their public websites; your system crawls those memos, indexes them, and answers users' legal questions with grounded, cited responses.

The prompt names three test areas: RAG + LLM + web crawling (crawling memos from law firm websites, and chunking them).

This round has a well-documented trap. Candidates walk in with deep LLM knowledge, spend the whole hour on retrieval and generation, and get rejected. The ingestion side, and specifically web crawler scale-up, is where interviewers push. This is a Software Engineer round, not an ML round: the interviewer wants pipelines, queues, scheduling, and failure handling, not embedding-model theory.

Treat the system as two planes of equal weight: an ingestion plane (crawl, parse, chunk, embed, index) and a serving plane (retrieve, rerank, generate). Budget your whiteboard time roughly evenly between them, and volunteer the crawler scale-up discussion before you are asked.

This walkthrough follows the Interview Framework and is sized for a 45 to 60 minute round.