A team is building a document search feature for a Foundry-hosted assistant: articles from the internal wiki need to be converted into vectors and stored in a vector database so the assistant can retrieve semantically similar passages at query time. Which type of model should generate those vectors?
Select an answer to reveal the explanation.
Short Explanation
Imagine you wanted to organize a giant pile of wiki articles so that anything about a similar topic ends up physically near anything else about that topic, even if the wording is completely different. That's what turning text into vectors is for: a model reads each passage and outputs a list of numbers that captures its meaning, positioned so similar meanings land close together. A model built for holding a conversation is tuned for a different job entirely, generating natural-sounding replies, not producing consistent coordinates you can compare mathematically. Trying to repurpose it for that would be like asking a chef to also do the restaurant's bookkeeping, technically capable of numbers, but not built for it. Tools that convert speech to text or generate images from prompts don't even operate on the same kind of input here, since the source material is already written text. What you actually want is a model whose entire purpose is producing those comparable numeric fingerprints, so the search system can find passages with related meaning fast.
Full Explanation
The correct answer is D. Embedding models are built specifically to convert text into fixed-length numeric vectors positioned in a space where semantically similar passages sit close together, which is exactly what a vector database needs to support similarity search over the wiki articles. Option A is incorrect because a chat completion model is optimized to generate conversational text responses, not to produce the consistent, similarity-optimized numeric vectors a search index depends on; using its output for search would be unreliable and inefficient. Option B is incorrect because a speech-to-text model converts audio into text and has no role here, since the wiki articles are already written text that needs to be vectorized, not transcribed. Option C is incorrect because an image generation model produces pictures from prompts; it has no mechanism for representing text meaning as a searchable numeric vector and is irrelevant to a text-based document search feature. Because the goal is to enable semantic retrieval of similar passages at query time, the team needs a model purpose-built to translate meaning into comparable numeric form, which is precisely the role of an embedding model in this pipeline.