DJames617 Service Comparison
RAG vs Fine-Tuning vs Prompt Engineering
Improve model behavior by changing instructions, grounding context, or model weights.
What They Have in Common
Each option can solve part of the scenario. The exam expects you to choose the one that satisfies the stated constraints with the right operational burden, security boundary, availability model, and cost behavior.
Key Differences
| Option | Primary model | Best use | Security | Availability | Operations and cost |
|---|---|---|---|---|---|
| Prompt engineering | Change instructions/context | Fastest, lowest overhead; no model training | Control prompt data, permissions, and logging | Depends on the selected model service | Lowest cost and operations; behavior changes without training |
| RAG | Retrieve authoritative context | Use current/private documents while preserving source updates | Secure source permissions, embeddings, retrieval, and KMS | Retriever and model availability must be designed | Indexing/retrieval cost; operate ingestion and evaluation |
| Fine-tuning | Update model behavior with training examples | Higher cost/ops; useful for style, task behavior, or specialized examples | Protect training data, artifacts, endpoints, and access | Deployment availability depends on endpoint architecture | Training and hosting cost; manage datasets, versions, and evaluation |
Typical Exam Clues
Practice After Studying
Return to the certification guide that includes this comparison and launch domain training from there.