AIF-C01 ยท Designing foundation model applications
Where can you store and search embeddings on AWS?
Amazon OpenSearch Service, Aurora, Neptune and RDS for PostgreSQL as places to store and search embeddings.
Key points
- A vector store holds embeddings and finds the nearest ones to a query vector.
- On AWS that can be Amazon OpenSearch Service (k-NN search, also full-text), Amazon Aurora PostgreSQL and RDS for PostgreSQL (the pgvector extension, embeddings next to relational data), and Amazon Neptune Analytics (graph plus vector index, used for GraphRAG).
- Amazon Bedrock Knowledge Bases can use several of these as its vector store.
Source: https://docs.aws.amazon.com/bedrock/latest/userguide/knowledge-base-setup.html
Practice questions
What does a vector store do in a RAG application?
- It generates the final natural-language answer
- It trains the embedding model on user questions
- It finds stored chunks whose embeddings are most similar to the query
- It encrypts prompts before they reach the model
Show answer
Answer: C. It finds stored chunks whose embeddings are most similar to the query
It stores embeddings and returns the chunks whose vectors are closest to the query vector. It does not generate text, and it is not a training service.
Which AWS services can store embeddings for vector search?
- Amazon Polly
- Amazon OpenSearch Service
- Amazon CloudFront
- Amazon Aurora PostgreSQL
- AWS Budgets
Show answer
Answer: B. Amazon OpenSearch Service, D. Amazon Aurora PostgreSQL
Amazon OpenSearch Service supports k-NN vector fields, and Amazon Aurora PostgreSQL can hold vectors through pgvector. Amazon Polly (speech synthesis), Amazon CloudFront (CDN) and AWS Budgets (cost alerts) are not vector stores.
A relational database such as Aurora PostgreSQL cannot store embeddings, so a dedicated vector database is always required for RAG.
Show answer
Answer: False
False. With the pgvector extension, Aurora PostgreSQL and RDS for PostgreSQL store embeddings and run similarity search. A dedicated store is a choice, not a requirement.
Practice 8 questions on this topic in Bitzu
Short lessons, instant explanations and spaced review of what you get wrong. Free on iOS and Android.
