1. PDF intake
Send a single patent PDF or a small set of representative documents.
Patent structure extraction
Patent structure extraction from patent PDFs.
Structured rows. Chemistry fields. Review-ready outputs.
456
structured rows in the anonymous chemistry benchmark
433
rows with structure-linked fields in that benchmark run
57.2
minutes total wall time in the public benchmark snapshot
What you get
2. Structure pass
Pull chemistry-heavy rows instead of leaving the result as screenshots.
3. Review output
Return files your team can inspect, filter, and QC.
Useful outputs
- Structure-linked and structure-heavy rows
- Normalized chemistry names and text-forward mentions
- Assay-linked context where evidence exists
Best for
- Chemistry-heavy patents with images, schemes, and dense tables
- Private pilot work after first-pass target search is narrowed
- Row-level outputs for medicinal chemistry or review workflows
- Teams that want evidence-linked outputs, not only OCR text
FAQ
Do you need a whole patent set, or can I send a single PDF?
One PDF is enough to start. A single chemistry-heavy patent can be a good trial case.
What kinds of fields can PatWinnow return?
Depending on the document, useful fields can include structure rows, normalized chemistry names, and evidence-linked assay or other bioactivity context.
Is structure extraction always the first step?
Usually no. It is often better to search first, filter the set, then run deeper extraction only on the strongest documents.
What should I send for a structure extraction trial?
Send one to several representative PDFs, the fields that matter most, and any review deadline.
Related proof pages
Benchmark
Structure extraction benchmark
See the public metrics for row counts, normalized-field capture, and runtime.
Sample output
Redacted chemistry-output schema
Browse example fields and review-ready row layouts without publishing source patents.
Case study
Patent PDF extraction example
Review how a representative PDF batch turns into structured chemistry rows and evidence traces.
Request a trial
Send a few representative PDFs and the output fields you care about.
If you already know which patents are high-value, start there. If not, we can scope the search step first and then move into deeper extraction.