Which foundation model (FM) is appropriate for converting private data into vector embeddings for storage in a database, when building a chatbot using Amazon Lex and Amazon OpenSearch Service?