Study Databricks GENAI-ASSOC Prompt Augmentation: key concepts, common traps, and exam decision cues.
This lesson is about turning the right evidence into the right model input. The exam expects you to know when prompts need more context, when they need clearer structure, and when the problem is really upstream or downstream instead.
| Need | Better first instinct |
|---|---|
| include key facts from user input | augment the prompt with those fields intentionally |
| include retrieved supporting context | build grounded prompt assembly |
| choose development tooling | pick a framework that fits the chain complexity, not the other way around |
| If the issue is mainly about… | Better first read |
|---|---|
| wrong answer format | prompt structure and output contract |
| unsupported facts | retrieval or context quality |
| chain complexity and orchestration | framework and chain design |
| inconsistent reasoning across steps | prompt assembly and stage boundaries |
| Trap | Better rule |
|---|---|
| adding more prompt text when retrieval is weak | fix retrieval or source quality first |
| choosing a framework because it is popular | framework choice should follow chain needs |
| mixing style control with grounding | output format and factual support are different concerns |
A chain has the right retrieved evidence, but the model keeps ignoring citation instructions and returning unstructured prose. Which layer should you inspect first?
Correct answer: A. If the evidence is already present, formatting and output-contract failures point first to prompt and chain logic.
Prompt-development questions usually start by classifying the problem layer. If the model has the right evidence but formats it badly, fix the prompt or chain structure. If the source coverage is poor, adding more prompt text will not solve it. Framework choice is about fit for chain complexity, not prestige. The weak answer usually tries to compensate for missing retrieval with prompt length.