Question

You are building a product-description generator that must follow a specific style. You have a Delta table containing 200 approved (product_attributes, description) example pairs. At request time, you want to select the 3 examples whose product category matches the incoming request and dynamically insert them into the prompt as few-shot demonstrations before calling a Foundation Model API endpoint. Which approach best constructs this prompt programmatically?

Accepted Answer

Query the matching example pairs at request time, format them into a string, and inject them into a parameterized prompt template variable (e.g., a {examples} placeholder) that is filled before invoking the endpoint.

Answer

Fine-tune the foundation model on all 200 example pairs so the style is baked into the model weights, eliminating the need to pass examples in the prompt.

Answer

Hard-code all 200 example pairs directly into a single static system prompt so the model always has every example available regardless of category.

Answer

Store the examples in a vector search index and retrieve them semantically, then concatenate the raw retrieved chunks after the model's response for post-hoc validation.