About using generative AI
This page describes how generative AI tools were used in preparing this site and the related materials, in line with the NLnet / NGI Zero policy on transparency of generative AI use.
Tools used
- Claude Opus 4.7 (Anthropic) — accessed through Claude Code (a CLI tool integrated with Visual Studio Code). Primary system.
- Google Gemini (web chat) — used as a secondary sanity-check.
What AI was used for
- Translating and tightening text. Croatian is the lead author’s first language. AI helped translate notes and analysis into English and to keep the writing within the limits of grant application forms.
- Looking things up. Verifying EU rules (ViDA), classifying B2B e-invoicing models country by country, and comparing with existing work (EN16931, France’s AFNOR XP Z12-013, Peppol). All factual claims were cross-checked against public sources.
- Restructuring older notes. Reorganising earlier internal documents into the form they are in here.
- Trademark and domain due diligence for the project name.
- Document review — flagging inconsistencies and overclaims.
Where the content comes from
- The author’s own work and his domain knowledge from more than 25 years building ERP and accounting software in Croatia.
- His earlier public analyses (the 2018 letter to the Croatian Ministry of Finance and the follow-up meeting with the Tax Authority; the 2025 analysis of the missing ERP-to-intermediary layer).
- Public sources (the European Commission’s eInvoicing country pages and other reputable sources for industry context).
Who is responsible
The lead author (Davor Geci, Winpis j.d.o.o.) is responsible. All the design choices, all the technical decisions, all the scope definitions — those are his, and he went through every output to check it. The AI helped write faster and in better English. It did not think for him.
The same applies to anything that comes later under funded work. AI tools will be used where they speed up engineering, but the author remains the one reviewing, deciding and taking responsibility for the result. Any substantive GenAI use in funded deliverables will be disclosed and logged in accordance with the relevant funder’s policy.