A software vendor's Foundry-hosted assistant must always respond in the company's specific documentation tone and format, and it must also quote the current price of each product, which changes weekly through automated updates in a product catalog. The team wants pricing to always stay current without retraining the model, while the tone and formatting stay consistent. Which combination of techniques should they use?
Select an answer to reveal the explanation.
Short Explanation
Split the problem into what actually needs memorizing versus what needs to stay live. Tone and formatting are things you want baked in permanently, so training the model on examples of that voice makes sense — it's a stable pattern the model can just absorb once. Prices, on the other hand, change every week, and nothing about training sticks around fresh; the moment you bake this week's numbers in, next week they're wrong again, and now you're stuck retraining on a schedule forever. The fix is to let the model fetch the live catalog at the moment someone asks, so the number is always whatever's true right now, no retraining required. Trying to get pricing accuracy purely from retrieved documents while skipping the trained tone risks inconsistent phrasing, and hardcoding prices into instructions just moves the update burden onto a person who now has to remember to edit that message every single week.
Full Explanation
The correct answer is C. Fine-tuning is well suited to teaching a model a consistent tone and formatting pattern because those are stable, repeatable characteristics the model can internalize once, while retrieval-augmented generation is well suited to pricing because it pulls the latest values from the live product catalog at the moment of each request, so prices stay current without ever retraining the model. Option A is incorrect because retraining the model every week to bake in that week's prices reintroduces the exact staleness problem the team wants to avoid between fine-tuning runs, and it adds unnecessary weekly cost and delay. Option B is incorrect because writing tone and formatting are properties of how the model generates language, and retrieval only supplies source text for the model to draw facts from; retrieved passages do not reliably constrain the model's own phrasing or structure, so tone consistency would suffer. Option D is incorrect because manually listing prices in the system message requires someone to update that message every time prices change, which is exactly the manual, retraining-like effort the team is trying to eliminate, and it also wastes context space and cost on every request.