Presenting at IPAI Toulouse
June 2026

On 10 June 2026, Entreacte will be presented at the International Patent & AI conference in Toulouse, France. This post gives a preview of what we'll be showing — the live demo, the technical results, and the roadmap for the next phase of development.

What we'll present

The presentation centres on the Inter-Omnia-Glypha method — the core claim of USPTO patent 19/384,146. We'll demonstrate the full pipeline live: uploading a document image through the API, watching the reversed binarization stage extract différance objects in real time, and seeing the resulting CA entropy, GNN edge structure, and script classification appear in the browser.

The benchmark results are the centrepiece. On the Derrida De la grammatologie benchmark — a 1517×2492px scan at 300dpi — the Entreacte pipeline detects 1,675 différance objects, constructs 17,831 GNN edges, and produces a CA entropy score of 0.618. At 150dpi, processing time is 1.37 seconds, comparable to Tesseract's 0.58 seconds while producing a richer structural representation.

The IntervalGlyphNet result

We'll also present the IntervalGlyphNet classifier — a result we're particularly excited about. By training a dual-channel CNN on the interval patterns between characters rather than on the characters themselves, we achieve 97.1% classification accuracy across seven script families: Latin, Arabic, CJK, Devanagari, Hebrew, Greek, and Cyrillic.

This result has a clear implication for the patent claim: the void between characters is not just computable — it is sufficient for classification. A model trained on intervals generalises across scripts without any language-specific training data. This is the Inter-Omnia-Glypha property: the method applies to any optical sensing of any marks in any medium.

Applications we'll discuss

The conference audience includes patent attorneys, AI researchers, and industry representatives. We'll present three application domains that are immediately relevant:

Document authentication. The CA entropy fingerprint provides a tamper-evident signature for any document. Legal contracts, medical records, patent filings — any document can be fingerprinted at submission and verified at any later point without embedded watermarks.

Ancient and damaged scripts. The interval-based approach is particularly powerful for damaged documents where characters are partially illegible. The spacing structure often survives degradation better than the characters themselves, enabling analysis of manuscripts and inscriptions that conventional OCR cannot process.

Multilingual document processing. Enterprise document workflows increasingly handle documents in multiple scripts. The Entreacte pipeline processes any script with the same code path, eliminating the need for language detection as a preprocessing step.

What comes next

After Toulouse, the roadmap has three priorities. First, expanding the IntervalGlyphNet training dataset with real document scans from Archive.org — moving from 97.1% on synthetic data to comparable accuracy on real digitised documents. Second, opening the Pro API tier for commercial licensing, with pricing designed for research institutions and enterprise document processing teams. Third, submitting the benchmark results for peer review.

If you are attending IPAI Toulouse and would like to meet, please write to founder@entreacte.net. We are particularly interested in connecting with researchers working on ancient scripts, bioimaging, and document forensics.