Question 6
Domain 4: Assembling and Deploying ApplicationsWhich decomposition choice usually makes a GenAI system easier to debug?
Correct answer: B
Explanation
Separating retrieval, prompt assembly, and generation into explicit steps creates clear boundaries between components, so failures can be isolated to one stage. This follows the debugging principle that smaller, well-defined steps are easier to inspect and test than a single opaque end-to-end process.
Why each option is right or wrong
A. Combine retrieval, prompt building, and generation into one opaque function
An opaque end-to-end function hides where retrieval, prompting, or generation failed.
B. Separate retrieval, prompt assembly, and generation into explicit steps
In a GenAI pipeline, making retrieval, prompt construction, and model generation distinct stages creates observable checkpoints, so you can inspect the retrieved context, verify the assembled prompt, and then test the output independently. That separation is what makes fault isolation practical: if the answer is wrong, you can tell whether the issue came from missing documents, bad prompt formatting, or the model’s generation step rather than a single opaque workflow.
C. Fine-tune before collecting baseline metrics
Baseline metrics should come first so later tuning changes can be measured meaningfully.
D. Use one giant chunk per document
Oversized chunks usually reduce retrieval precision and make context problems harder to inspect.