Whenever you're ready, walk me through how you'd approach this.
How would you approach building a RAG (Retrieval-Augmented Generation) pipeline?
Explain the complete practical pipeline from document ingestion through retrieval and response generation. Cover document parsing, chunking, metadata, embeddings, vector storage, hybrid search, reranking, prompt construction, freshness, access control, evaluation, observability, failure recovery, and backfilling. Discuss how you would make ingestion idempotent, handle document updates and deletions, prevent stale or unauthorized context, and measure retrieval and answer quality. Include the technologies you would select, key data models, orchestration approach, and representative implementation details.