Whenever you're ready, walk me through how you'd approach this.
How do you optimize embeddings and vector search performance when scaling to millions of documents for TSMC internal retrieval?
