Matching a model task to a given business requirement

For a given business requirement, the exam expects you to match it to the correct model task type (e.g., text generation, summarization, extraction, classification, embedding/retrieval, or translation) rather than defaulting to a single general-purpose LLM approach. Choosing the right task type drives downstream decisions like model selection, evaluation metrics, and pipeline architecture (e.g., RAG vs. fine-tuning vs. simple prompting).

1 · Learn the must-know

• 1Question answering over private/proprietary documents typically requires retrieval-augmented generation (RAG) using embeddings and a vector search index, not just a base LLM.
• 2Semantic similarity, clustering, or document search tasks require an embedding model, not a text-generation model.
• 3Summarization and extraction tasks can often use smaller, cheaper, or specialized models rather than the largest general-purpose LLM, which is more cost-effective for production.
• 4Classification tasks (e.g., sentiment, intent, routing) may be solved with either a fine-tuned traditional ML model or an LLM via prompting, depending on latency, cost, and accuracy needs.
• 5Multi-step business requirements (e.g., summarize then classify then respond) often require chaining multiple model calls or tools rather than a single prompt.
• 6Selecting a task that is over-powered for the requirement (e.g., using a large generative model for simple classification) increases cost and latency without improving business outcomes.

2 · Check your understanding