A small regression test for Chinese and Japanese document ingestion: catch whitespace-based chunking, add a code-point cap, and preserve source offsets without confusing characters with embedding tokens.
Tag: RAG
How to Search Confidential PDFs Locally on a 12 GB Laptop—Without Overbuilding RAG
A practical, privacy-first path from searchable PDFs to local retrieval: extract text, measure full-text search, add OCR or embeddings only when the evidence says you need them, and keep every result tied to its source.