Design a Deep Research Agent

Design an agent that takes a hard research question, searches and reads dozens of web sources in parallel, and writes a report in which every claim cites a source it actually fetched.

Multi-AgentPlanningWeb ToolsCitationsBudgetsContext Management

Key Requirements

Plan sub-questions and run searches in parallel
Keep a per-report budget in dollars, searches and time
Save each finding with a verbatim quote and its URL
Check every citation before the report ships
Stop on coverage, and name gaps and conflicts

Interviewer Follow-ups

Q1How do you handle contradictory sources?
Q2How do you cap cost and latency for a 50-source deep dive?
Q3How do you prevent hallucinated references?

Design a Text-to-SQL / Conversational Analytics Agent

Design an LLM Gateway / Model Router & Inference Proxy