Volume 7

Volume 7 - Retrieval-Augmented Generation (RAG)

Volume 7 designs a retrieval-augmented knowledge assistant for an illustrative company of 5,000 employees and prices every stage of it. Its 150,000 documents become about a million searchable chunks, its 12,000 questions a day share a 1.5-second p95 budget for the first token, 30% of its content is restricted by department and role, and an edited policy must be answerable within 15 minutes. Section 1 builds the index, Section 2 decides which evidence reaches the prompt and who may see it, and Section 3 takes the assistant to production, closing with a design review whose model answer costs $5,643.91 per 30 days at September 2026 list prices.