Prepare PDF for AI
Convert PDFs into chunked JSON documents ready for RAG pipelines, embeddings and LLM ingestion.
Upload PDF Files
關於此工具
Prepare PDF for AI turns documents into the JSON shape retrieval pipelines expect: an array of { text, metadata } records compatible with LlamaIndex and LangChain loaders.
Three chunking strategies are available. Per-page keeps citations exact, fixed-size character windows with overlap suit embedding models with a token budget, and whole-document is best for short files. Chunk boundaries prefer sentence breaks so meaning is not cut mid-thought.
Every chunk carries its source file, page number and document metadata, and batches of PDFs come back as a ZIP. Nothing is uploaded anywhere.
使用方法
Upload Your PDFs
Add one or many PDFs — they are processed together.
Pick a Chunking Strategy
Choose per-page, character windows or whole-document, then set chunk size, overlap and a minimum length.
Extract and Download
Click Extract and download the JSON, or a ZIP when processing several files.
使用場景
Build a RAG Index
Produce clean, page-attributed chunks to embed into a vector store.
Chatbot Knowledge Bases
Convert manuals and policies into retrievable documents.
Citable Answers
Keep page numbers on every chunk so generated answers can cite their source.
常見問題
Which chunking strategy should I use?
Per-page for citation accuracy, character windows when your embedding model has a strict token limit, whole-document for short files.
Why does overlap matter?
Overlap keeps a sentence that straddles a chunk boundary retrievable from either side, which measurably improves recall.
Does this work on scanned PDFs?
No — it reads the text layer. Run OCR first.