Study Databricks GENAI-ASSOC Business Requirements and Chains: key concepts, common traps, and exam decision cues.
Many weak GenAI designs fail before the first prompt is written. The exam checks whether you can read a business requirement, identify the real task, and define what the application should take in and return before reaching for tools.
| Design question | Why it matters |
|---|---|
| what input does the user provide? | shapes retrieval, prompt assembly, and safety checks |
| what output does the system need? | determines the model task and response format |
| does the app need retrieval, tools, or actions? | decides whether a simple prompt, chain, or agent pattern fits |
| what is the real business goal? | prevents solving a different problem elegantly |
| Need | Better first instinct |
|---|---|
| rewrite text in a constrained format | text generation or transformation |
| classify an intent or route a workflow | classification or decision-oriented task |
| answer with grounded source evidence | retrieval-backed generation |
| take action or use multiple tools | agentic or tool-using chain pattern |
| If the business requirement emphasizes… | Strong first read |
|---|---|
| exact output format and reviewability | define the output contract before you choose the model or framework |
| proprietary knowledge and citations | design retrieval into the chain early |
| multi-step decision flow or tool usage | map the stages explicitly instead of calling everything a chatbot |
| low operational risk | prefer the simplest chain that satisfies the requirement |
| Trap | Better rule |
|---|---|
| starting with a framework instead of the requirement | define the business task first |
| overbuilding an agent for a single-step job | use the least complex design that fits |
| describing outputs vaguely | the exam rewards concrete output definitions |
A team wants an assistant that takes a support ticket, checks internal policy documents, and returns a structured escalation recommendation with citations. What should you define first?
Correct answer: B. This exam rewards starting with the business contract and only then selecting retrieval, tooling, or model patterns.
Design questions usually reward solving for the business contract before the tooling. First, define the required input and output clearly. Second, decide whether the answers must be grounded over enterprise content, structured into exact fields, or just conversational. Third, add prompt, retrieval, or chain design only after the contract is clear. The weak answer usually starts with model or framework hype.