Skip to content
Few-Shot Prompt Builder
Turn example input/output pairs into a clean few-shot prompt for classification, extraction, rewriting or formatting tasks with any LLM.
🔒 Processed locally in your browser — we don’t upload your input.
How to use the Few-Shot Builder
Few-shot prompting means showing a language model a handful of worked examples before asking it to handle a new case. It is the quickest way to teach a model a specific labelling scheme, extraction format or writing style without any fine-tuning. This builder lets you write an instruction, paste example pairs, and then add the new input; it formats everything into a consistent pattern the model can copy.
Enter one example per line using a separator such as “=>” between input and output, for example “I love this phone => positive”. You can rename the Input and Output labels to match your task (Review/Sentiment, Email/Category, Sentence/JSON) and choose between plain-label or XML-tag formatting. XML tags are useful with Claude and other models that respond well to clearly delimited sections, while plain labels are compact and readable.
Good few-shot sets are short, varied and correct: three to eight examples covering edge cases usually beat dozens of near-duplicates. When your examples grow into hundreds, that is the point to consider fine-tuning instead, and you can convert the same pairs into training data with the JSONL Dataset Builder. Everything stays in your browser.
Frequently asked questions
How many examples should I include?
Typically 3–8 diverse examples. More examples cost tokens and give diminishing returns.
What separator should I use?
Use => by default, or change it to a tab, | or any string that does not appear in your data.
What is the difference from fine-tuning?
Few-shot examples live in the prompt each time; fine-tuning trains them into a model permanently.
Is my data uploaded?
No. Formatting happens locally.
About this tool
Turn example input/output pairs into a clean few-shot prompt for classification, extraction, rewriting or formatting tasks with any LLM. People also look for: few shot prompt, few-shot examples, in-context learning, prompt examples.
Related searches