RAG Chunking Prompt Pack

You are preparing a document for a retrieval system. Split the text below into self-contained chunks. RULES - Target 200-350 words per chunk. Never split a sentence, list, table, or code block across chunks. - Prefer to break at heading and paragraph boundaries. - Each chunk must make sense on its own. If a chunk depends on a heading above it, restate that heading inside the chunk. - Do not summarize or rewrite the source. Preserve wording exactly. OUTPUT: a JSON array of objects: { "chunk_id": <int>, "heading": <string|null>, "text": <string> }. DOCUMENT: This is a sample document for demonstration purposes. It has multiple sections. Section one has two paragraphs. This is the first paragraph. It has a list: - item one - item two The second paragraph is here. Section two has a table: | column one | column two | | --- | --- | | cell one | cell two | The document ends here. [{"chunk_id": 1, "heading": "Section one", "text": "This is a sample document for demonstration purposes. It has multiple sections. Section one has two paragraphs. This is the first paragraph. It has a list: - item one - item two"}, {"chunk_id": 2, "heading": "Section one", "text": "The second paragraph is here."}, {"chunk_id": 3, "heading": "Section two", "text": "Section two has a table: | column one | column two | | --- | --- | | cell one | cell two | The document ends here."}] For the chunk below, write a single sentence that situates it within the whole document: what section it is from and what it is about. Return only that sentence. WHOLE-DOCUMENT SUMMARY: This is a sample document for demonstration purposes. CHUNK: This is a sample document for demonstration purposes. It has multiple sections. Section one has two paragraphs. This is the first paragraph. It has a list: - item one - item two From section one, this chunk introduces the document and its sections, including a list of items. For the chunk below, extract retrieval metadata. Return JSON only: {"title": "Section two", "keywords": ["table", "cell"], "entities": [], "content_type": "table", "answers_questions": []} CHUNK: Section two has a table: | column one | column two | | --- | --- | | cell one | cell two | The document ends here. From the chunk below, write up to 3 question/answer pairs a real user might ask, where the answer is fully supported by the chunk. If the chunk cannot support a grounded question, return an empty array. Never use outside knowledge. CHUNK: This is a sample document for demonstration purposes. It has multiple sections. Section one has two paragraphs. This is the first paragraph. It has a list: - item one - item two [{"q": "What sections does the document have?", "a": "multiple sections"}, {"q": "How many paragraphs are in section one?", "a": "two paragraphs"}]