Design a RAG system with agentic tools
Quick Overview
This question evaluates a candidate's ability to design a Retrieval-Augmented Generation (RAG) system with agentic tool-calling for enterprise knowledge bases, testing competencies in scalable ML system architecture, data ingestion and indexing, retrieval and reranking, prompting, tool integration, evaluation, monitoring, and safety (ML System Design domain). It is commonly asked to assess architectural reasoning, trade-off analysis, and handling of evolving document stores for grounded QA, combining practical application (system design and operational considerations) with conceptual understanding of retrieval, generation, and guardrails.
Design a Retrieval-Augmented Generation (RAG) question-answering system for an enterprise knowledge base.
Requirements:
- Users ask natural-language questions; the system answers with grounded responses and citations.
- The knowledge base includes documents that change over time (updates, deletions).
- The system should handle multi-step questions, and may use agentic tool-calling (e.g., search, calculator, database lookup).
- Discuss architecture, data ingestion/indexing, retrieval and reranking, prompting, tool use, evaluation, monitoring, and safety/guardrails.