Core Architecture

Most AI PDF tools convert your file into plain text before analysing it. In doing so they destroy fonts, tables, diagrams, and layout. The extracted output is a degraded approximation of the original.

This system takes a different approach. The AI reads your document to make a decision — which pages to keep. Once that decision is made, the AI is done. The engine goes back to the original raw PDF file, slices the approved page data byte-for-byte, and writes it into a new file. No rendering, no conversion, no re-encoding.

When multiple files are uploaded in a single job, each file is extracted independently in parallel. The AI analyses all files simultaneously, and the resulting keep/discard verdicts are merged into a unified ordered list before the output file is assembled.

The Study Editor stores files locally using IndexedDB — a low-level key-value store built into every modern browser. Files persist between browser sessions and survive page refreshes.