A logistics company wants a Foundry-hosted assistant that can look at photos uploaded by drivers of damaged packages and generate a written damage description for insurance claims. Which type of model should they deploy for this capability?
Select an answer to reveal the explanation.
Short Explanation
The task here is to look at a picture and describe what is wrong with it, so whatever model gets deployed actually has to be able to see the photo, not just process words about it. That means reaching for a model built to accept images alongside text and reason about both together. An embedding model that only handles text can turn words into searchable vectors, but handing it a photo of a dented box gives it nothing it can work with. Bumping up to a bigger text-only chat model does not fix that either — size does not grant the ability to perceive pixels, and there is no way to infer damage from a file name alone. A speech-to-text model is built for a different job entirely, turning spoken audio into text, which does not help when the input sitting in front of you is a photograph.
Full Explanation
The correct answer is C. Analyzing photos of damaged packages and producing a written description requires a model that can accept image input alongside text, which is exactly what a multimodal model with vision capability provides — it can interpret the visual content of the photo and generate the corresponding text description. Option A is incorrect because a text-only embedding model converts text into numeric vectors for tasks like search and similarity comparison; it is not designed to interpret the visual content of a raw photograph. Option B is incorrect because a larger text-only chat model still has no way to perceive an image's content just because it is bigger — model scale does not substitute for the ability to process visual input, and a file name carries no information about what the photo shows. Option D is incorrect because a speech-to-text model transcribes spoken audio into text; it addresses a completely different input modality and would not process an uploaded photograph at all.