Axion Playground · by Iris.ai Your AI is only as powerful as the context you give it.

Axion turns fragmented enterprise documents into the structured, trusted context your AI projects need. Here it processes 14 patent filings across six languages, five scripts and seven patent offices, including 157 scanned pages with no text layer at all. What comes out is one structured view of what each material is made of, how it is made, and what properties result. Same pipeline for every filing, no tuning per document. About seven minutes, and you set the pace.

~7 minutes 14 patents · 376 pages 6 languages · 5 scripts

The dataset

What Axion is reading

A portfolio of public patent filings, assembled to match the shape of a real R&D and IP portfolio. You follow it from raw files to finished output.

DEMO

Materials and manufacturing patent portfolio

14 filings on steel, coatings and process manufacturing: Zn-Al-Mg coatings, electrical steel, duplex stainless, ultra-high-strength automotive plate. Seven patent offices, filed between 2012 and 2025. All public filings, no customer data.

  • Six languages across five scripts: Chinese, Japanese, Arabic, Korean, French, English
  • 157 of the 376 pages are scanned images with no text layer, so every character is read off the page image
  • Most of the substance sits in composition tables and technical drawings. Very little of it is in prose
6 languages 5 scripts 7 patent offices 42% scanned
✦ Start Demo
⤓  Use your own filings

[Data handling copy pending security sign-off.]
Bring your own patent filings and we run the same pipeline on them.

The finish line

What you'll have when you finish

Three things, all of them real output from these 14 filings.

01

One pipeline across six languages and five scripts

A Japanese electrical-steel filing, a scanned Arabic filing and a French duplex-stainless filing go through the same path and come out in the same shape. Nothing was tuned per document.

Hanzi · Kana/Kanji · Hangul · Arabic · Latin
02

Findings connected across a figure, a table and the text

When a filing states the same result in more than one place, Axion links those mentions, so you can see where a number came from. 52 connected findings here, across 11 of the 14 documents.

Every row traces back to a page
03

A structured view of the whole portfolio

What each material is made of, how it is made and what properties result, lined up across 14 filings. Alongside it, the original page with every element boxed and labelled, so you can check any row against the source.

Charts rebuilt as data, not screenshots
126tables
95figures
54charts
52connected findings

337 artifacts out of 376 pages. Where a person still looks: continuous-curve charts such as spectra and distribution curves reconstruct with lower confidence than discrete bar and point charts, and Axion shows that confidence instead of hiding it. You spot-check the fields that matter to you, rather than reading 376 pages in six languages.

After the demo

By the end, you'll know whether Axion handles documents like yours.

If it does, request a pilot and we'll run it on your own filings. If the timing is off, keep the material and come back.

Request a pilot Start the demo first

You can come back to the playground any time.