Patron GO, the Prague-based "financial antivirus" app run by Antivirus s.r.o., is rebuilding its recommendation engine around a self-learning AI core in partnership with BigHub and the Faculty of Nuclear Sciences and Physical Engineering at Czech Technical University in Prague (FJFI ČVUT). The app connects to users' bank accounts in read-only mode via PSD2 and looks across the household budget, from bank fees and insurance to energy and mobile tariffs, suggesting cheaper alternatives and rewarding users for better money habits. It serves more than 600,000 users.
The motivation was scale. Every transaction had to be matched to the right merchant and category, and recommendations came from hand-written business rules. With hundreds of thousands of users, manual categorisation became a bottleneck: new merchants and payment descriptions appear daily, rules could only give generic advice such as "your energy tariff is above average", and each new country would have required rebuilding the mappings and rules from scratch.
Patron GO 2.0 has four components. Machine-learning models categorise every transaction automatically, replacing manual rules. Community self-labelling lets users confirm or correct categories, and the models retrain on those labels, so new markets can be mapped by local users instead of local teams. Benchmarking groups users by financial behaviour so each sees how their spending compares with similar people. Generative AI turns those insights into personalised recommendations and financial-health explanations that can be localised for new languages. The stack runs on Google Cloud with MLOps for production model integration.
Roles are split three ways: Patron GO owns the product, platform, scaling and PSD2 data security; BigHub builds the AI core architecture, data preparation, classification models, the self-labelling system and production integration; FJFI ČVUT researches self-learning models, transfer learning and model portability to new markets.
The project runs from 1 June 2025 to 31 May 2027 and is co-funded by the European Union under the Czech Operational Programme Technologies and Applications for Competitiveness (OP TAK, reg. no. CZ.01.01.01/01/24_062/0007593). The AI modules are being piloted with a selected group of Czech users and tuned for categorisation accuracy, personalisation and offer relevance; BigHub states that measured results will be published after the pilot. Next steps are localising models and adding PSD2 connectors for Slovakia and Poland, then rolling the AI-driven app out to all users in Czechia and the pilot markets.