AI Persona Generator
Create a persona for a chatbot or roleplay character, as a prompt or a character card.
Turn input and output pairs into a clean, consistent few-shot prompt block.
The Few-Shot Example Builder turns input and output pairs into a consistently formatted few-shot block — XML tags, labels, markdown, chat messages or JSONL — and checks the examples for the inconsistencies that make answers unpredictable.
Few-shot prompting means showing the model a handful of worked examples before the real input. It is the most reliable way to control output format, label names and tone — more reliable than instructions alone — and it only works if the examples are formatted consistently.
Paste your examples as Q: and A: pairs, two tab-separated columns from a spreadsheet, or blocks separated by blank lines. The builder formats them all the same way: XML tags, which Anthropic recommends for Claude; Input and Output labels; markdown headings; a chat messages array where each example becomes a user and assistant turn; or JSONL pairs. An empty slot for the real input can close the block.
It also checks the examples themselves. Outputs in different shapes — some JSON, some prose — are the most common reason a few-shot prompt gives inconsistent answers, and a single example tends to be copied rather than generalised from. Both are flagged, along with the token cost of the block.
XML tags, labels, markdown headings, a chat messages array or JSONL pairs.
Q: and A: labels, tab-separated spreadsheet columns or blank-line blocks are all read.
Outputs in mixed shapes and very uneven lengths are flagged before they confuse the model.
Replace Input and Output with labels that describe your data, such as Ticket and Category.
See what the block adds to every request before you commit to it.
Q: and A: labels, tab-separated columns or blank-line blocks are all detected.
XML tags for Claude, labels or markdown for most models, chat messages for an API.
Use labels that describe the data, such as Ticket and Category, rather than Input and Output.
Paste it into your prompt, with the real input filling the slot at the end.
Usually two to five. One example tends to be copied rather than generalised from, and past five or six the gains flatten while every call gets more expensive.
Both work. Chat-message examples mirror a real conversation and suit chat APIs; inline examples in XML tags or labels are easier to read and edit in a single prompt.
The model learns the pattern across the examples. If one output is JSON and another is a sentence, it will mix the two in its answers.
When you need hundreds of examples to capture the task, or when the examples make every request expensive. Few-shot is the right first step because it costs nothing to try.
No. Formatting happens in your browser.
What people who use the Few-Shot Example Builder think of it. Reviews come from visitors and are posted as written.
No ratings yet
Yours would be the first.
0 reviews
Newest first
No reviews yet
Be the first to rate the Few-Shot Example Builder. It takes half a minute and no account.
Write the first reviewCreate a persona for a chatbot or roleplay character, as a prompt or a character card.
Build a Stable Diffusion negative prompt from the problems you keep getting.
Find the words and habits that make text read as AI-generated, so you can edit them out.
Convert a conversation into ChatML, Llama 3, Mistral, Gemma or Alpaca prompt templates.
Generate MCP server config for Claude Desktop, Claude Code, Cursor, VS Code and Windsurf.
Score a prompt on the parts that change the answer, and see what is missing.
Add aspect ratio, stylize, chaos, style references and more to a Midjourney prompt.
Turn a rough idea into a structured prompt that gets a usable answer.