A team wants to make PDFs and web pages searchable by meaning for a retrieval-augmented generation assistant. The pipeline must extract the text, split it into passages, and produce vectors that a retriever can search. Which combination of Google Cloud capabilities fits?
Show answer and explanation
Answer: D
A RAG ingestion path first converts PDFs and web pages into text with a document parser such as Document AI, then splits that text into chunks, and generates an embedding for each chunk so a vector index can retrieve passages by meaning. Each stage has a clear role. Full-text indexes match keywords rather than producing semantic vectors, regression models do not parse documents or create retrieval embeddings, and raw bucket storage leaves the content unprocessed for search.
Why the other options are wrong
- A. Full-text search matches words and phrases, but it does not create semantic vectors, so questions phrased differently from the source text would miss relevant passages.
- B. Regression models predict numeric targets from structured features, so training them on document text does not produce passages or vectors that a retriever can search.
- C. Storing PDFs as-is gives the retriever unparsed binary content, so it cannot find meaningful passages without the extraction, chunking, and embedding steps.