LiteParse Node-only PDF parser ported to browser in an hour via Claude Code

A Node-only PDF parser got a working in-browser version basically overnight, so you can just drop files into a web app instead of wiring up a CLI. The demo handles messy layouts like multi-column PDFs and shows the structure directly.

Under the hood, LiteParse (from LlamaIndex) skips ML and maps text onto a monospace grid to preserve layout, which is why it stays fast enough for agent workflows. It already supports 50+ formats, and its ParseBench benchmark covers 568 real-world enterprise pages. The browser build and notes came out after about an hour-long Claude Code run.