Retrieval augmented generation that grounds responses: Which deployment technique best satisfies these
RAG grounds answers in the newest internal documents at query time without the recurring cost of fine-tuning.
The question
A firm wants a general model to answer employee questions using its constantly-updated internal policy library, cannot afford repeated fine-tuning runs, and needs answers to reflect the newest documents the day they change. Which deployment technique best satisfies these constraints?
Preparing for AIGP? Take the free 5-min readiness quiz →
- Periodic full fine-tuning of the base model on the internal policy corpus each time documents changeRepeated fine-tuning is exactly the recurring cost the firm cannot afford and lags behind daily document updates.
- Deploying the base model as-is and relying on its pretrained knowledge to answer the policy questionsAn as-is model has no access to the firm's private, changing documents and would answer from stale generic knowledge.
- Retrieval augmented generation that grounds responses in the current internal documents at query time ✓RAG pulls the latest documents at inference without retraining, so answers stay current and costs stay low, matching every constraint.
- Building an agentic architecture that autonomously executes multi-step workflows across internal systemsAgentic orchestration adds action-taking complexity the use case does not need and still lacks a freshness mechanism.
The trap
Defaulting to fine-tuning for domain knowledge when the data changes constantly and retraining budget is limited. How to remember it
RAG grounds answers in the newest internal documents at query time without the recurring cost of fine-tuning.
How many of these would you get right?
One of 1581 AIGP questions on Certsqill. Take a free five-minute check and see your score per domain — not one number, but which section to open tonight.
Test your AIGP readiness — freeMore Understanding How to Govern AI Deployment and Use questions
- An open-source model whose weights can be self-hosted: Considering model-type differences, which option best →
- Performing an impact assessment that evaluates effects: Which activity is being described? →
- How the vendor may use customer input data and who is: When reviewing a third-party AI vendor's licensing →
- All 424 Understanding How to Govern AI Deployment and Use questions →
Part of the Certsqill AIGP question bank · Understanding How to Govern AI Deployment and Use ·
Every answer, right and wrong, comes with its own explanation.
