Question 18
Domain 4: Assembling and Deploying ApplicationsA support chatbot needs to answer using a frequently changing internal knowledge base. The team wants to avoid retraining the language model after every document update. Which design is the best fit?
Correct answer: A
Explanation
RAG fits because the exam guide says Databricks supports “performant RAG applications” and “Vector Search for semantic similarity searches,” which lets the app pull fresh answers from an indexed source instead of retraining. Using a governed document source also matches “Unity Catalog for data governance,” so updates can be managed centrally as the knowledge base changes.
Why each option is right or wrong
A. Use retrieval-augmented generation with an indexed, governed document source
The exam guide explicitly frames Databricks as supporting “performant RAG applications” and “Vector Search for semantic similarity searches,” which is the right architecture when the knowledge base changes frequently and you need fresh retrieval at query time rather than model retraining. Because the source content is governed in Unity Catalog, updates can be managed centrally while the retriever continues to query the indexed documents without any fine-tuning cycle or retraining delay.
B. Fine-tune a model every time a document changes
Fine-tuning is for model adaptation, not frequent document refreshes in a changing knowledge base.
C. Ask users to paste the entire knowledge base into each prompt
Prompts have limited context and are not a practical way to carry an entire knowledge base.
D. Ignore internal documents and rely only on model pretraining
Model pretraining is general knowledge, not a substitute for internal, up-to-date documents.