About PaperInsight

What it does

PaperInsight turns a research PDF into a structured report: the problem the paper addresses, why that matters, the method, the results, the limitations, and a plan for reproducing the work. Each section names the pages it was drawn from.

It is not a chat window over a PDF. The output is a fixed structure you can read top to bottom, or come back to later at the same link.

How it works

The PDF is read page by page and split into passages that keep their page numbers. Those passages are grouped by the sections the paper actually has, and each group is summarised in turn. Those summaries are then synthesised into the report.

Reading the paper in stages rather than sending the first few thousand characters is what lets a forty-page paper's limitations get the same attention as its abstract.

Every generated report is checked against a fixed schema before it is saved. If it does not validate, one repair pass runs; if that fails too, the job fails and says so. A report you can read is a report that passed.

What to be careful about

The reports are generated by a language model. The prompts instruct it to use only the supplied text — not to infer claims, invent numbers, or attribute limitations the paper never states — and the page references let you check. Use them. Treat the report as a way into the paper, not a replacement for it.

Scanned or image-only PDFs cannot be read yet: there is no OCR. Papers whose structure is unusual may have sections detected imperfectly.

Your documents

Uploaded PDFs are stored so the report can be rebuilt and the source passages shown. Analysis runs on a locally hosted model by default — nothing is sent to a third-party AI provider. You can delete a paper and everything derived from it at any time.