prompt-engineering

Include 3-5 input-output example pairs covering diverse cases and edge cases before the target input. For sentiment classification, show labeled examples of positive, negative, and neutral reviews, then append the new review for the model to classify.

Chain-of-thought prompting asks the model to reason step-by-step before answering, improving accuracy on math and logic tasks. Use it for multi-step reasoning problems; self-consistency variants generate multiple reasoning paths and pick the most common answer.

Specify the exact JSON schema in the prompt with field names and types, then show an example response. For production use, OpenAI function calling enforces structured outputs by defining parameter schemas the model must populate.

Measure accuracy against labeled test cases, consistency by running the same prompt multiple times, and cost from input and output token counts. A/B test prompt variants on a test set to compare accuracy against per-request cost.

Use ReAct when the task requires external tools like search, calculators, or database lookups. It interleaves Thought, Action, and Observation steps so the model reasons about which tool to call and incorporates results into its answer.