Exercise: Design RAG for customer support
Requirements: Answer questions from K support articles. ms latency target.
Your design:
Chunking: Recursive, tokens, overlap. Keep article metadata.
Embedding: OpenAI text-embedding--small (fast, good quality).
Vector DB: Pinecone (managed, low-latency).
Retrieval: Hybrid search (articles have product names). Retrieve top-.
Generation: Include source links. Cite article titles.
Follow-up questions: How do you handle updates? How do you measure quality?