Skip to content
ML Academy

RAG and agents

Interview question

  • seniorlevel
  • 15 minsuggested time
  • 2Applied AI, ML engineer

A research agent for an airline's network-planning team calls a search tool, a SQL tool and a spreadsheet tool in a loop, up to eight steps per task. In testing it finishes 66% of tasks, costs several times more than the single-prompt prototype, and sometimes repeats the same call. Each step is "95% reliable". How do you explain the numbers and what do you change?

Tests: Reasoning about compound reliability, cost growth and loop control in agentic systems; deciding workflow vs agent.

day 49day 50day 54day 52

Sign in to practise: answers are saved before model answers are shown.