In cooperation with R&D experts from Norway, the team of the Lithuanian start-up FINBRO managed to significantly improve its financial services platform. During the project, the classification models for legal and natural persons were being developed, tested, and evaluated under real conditions. The work that was being performed until then was also continued: data preparation, collection, summarization, model testing.
In working with „Norway Grants“’ support and in cooperation with Anders Stølan and Ivelin Andreev, the R&D specialists, the FINBRO team managed to improve notoriously within a few months the product they were developing – a smart system that examines the parameters of financiers and at the same time assesses the needs of users according to the relevant criteria. So, later the users can only see the financiers that are relevant to them and their comparison is allowed. Such a system saves a lot of time for both parties.
In the first phase of the project, the FINBRO team started developing its loan comparison platform: various internal and private data were collected, analysed, and categorized. It is therefore natural that, in the second phase of the project, priority was given to elimination of the software deficiencies observed during testing. The system algorithms were also improved: calibration of models was performed, and their parameters were updated.
Formation of the overall view of the system was started during the work. Creation of synergy between the external user interface and that which is happening inside the smart platform was also started. Training of the model for classification of legal entities’ applications to financiers and also optimization of hyperparameters were performed inside the system using the loss functions that were tested during the project, which were used as the optimization evaluation function.
Meanwhile, model classification user interface was created externally, that is, in the product part that will be visible to the end user. The system allows the user to submit an application, then various registers are collected (financial statements, credit reference reports, and other data). All this is converted into a json file which is examined by artificial intelligence (AI). It intelligently distributes the application to financiers and it is, therefore, received only by such financiers for whom the application is relevant. In developing the AI of the system, FINBRO received significant help from Židrina Pabarškaitė, who is an expert in artificial intelligence. She has studied the peculiarities of artificial intelligence not only in the universities of Lithuania, but also of the United Kingdom, and has also worked with many AI companies.
Already at this stage, the FINBRO team considered the possibility of testing their system with a wider group of users, but this was not achieved. Although there was communication with the Bank of Lithuania regarding the use of the test environment, the answer received was that the Bank of Lithuania has not personalized the MADB data and, therefore, they cannot invite market participants to try the trial version.
So, the classification models for lending companies and natural persons were tested in laboratory conditions. Nevertheless, within a few months, the FINBRO system was significantly improved during the course of the project and acquired its final form.
“We are happy not only about the financial support from „Norway Grants“, but also about the great assistance of Norwegian R&D specialists. All this allowed us to improve our system much faster than planned. Now that, due to the growing interest rates of central banks and due to stricter rules for the provision of financial services, loan comparison platforms are becoming more and more necessary, speed becomes of the essence,” – summarizes Mantas Norvaišas, founder of the FINBRO platform.
Project No LT07-1-EIM-K02-006 “UAB Finbro R&D operations in the sphere of ICT to boost competitiveness” is financed by the Norwegian Financial Mechanism 2014-2021 and the state budget of the Republic of Lithuania.