“I need to extract structured data from unstructured text using an LLM. Walk me through writing a few-shot prompt that reliably returns valid JSON, including how to handle edge cases where the information is missing from the input.”
“I’ve been using zero-shot prompts and getting inconsistent results. Can you help me convert this prompt to a few-shot prompt? Here’s my current prompt: [paste your prompt]. Here are three examples of inputs and the outputs I’d consider correct: [paste examples].”
“Explain chain-of-thought prompting to me with a concrete example relevant to data processing or business logic — not maths puzzles. When would I choose chain-of-thought over just a better-written direct prompt?”
“Write me a Python function that tests an LLM prompt against a list of test cases and reports: (1) whether the output is valid JSON, (2) whether all required keys are present, (3) whether constrained fields contain only valid values.”
“I’m building a prompt that needs to produce consistent output across many different inputs in production. What are the five most important prompt engineering practices I should follow to maximise consistency?”
