3. Vector Stores, RAG, and Prompt Engineering (Domain 1 continued)

The concepts in this phase carry disproportionate exam weight because they appear in scenarios for every other domain. Vector store architecture decisions affect security (Domain 3), cost (Domain 4), and evaluation (Domain 5). Prompt engineering decisions affect everything. A weakness here cascades across the entire exam.

Phase 3 goes deep on the mechanics that most candidates only understand superficially — not just "what is RAG" but why specific chunking decisions break retrieval, why hybrid search outperforms pure vector search for certain query types, and why system prompt governance is a security control, not just a quality control.

⚠️ Common Misconception: Vector stores are just databases with a special search feature. A production vector store is an architecture decision that affects ingestion throughput, retrieval latency, metadata filtering capability, cost per query, and update frequency. The choice of OpenSearch vs. Aurora pgvector vs. Bedrock Knowledge Bases' managed store locks in trade-offs across all these dimensions simultaneously.