Robot Emma (Electronic Multimedia Assistant) helps to improve marketing efforts.

A kratt for analysing conversations with customers at The Estonian Health Insurance Fund.

The process of keywording documents is a resource-intensive process. It relies on the knowledge and assessment of individuals and it is not always possible to guarantee the consistency of keywords across different versions of the same publication. This is manual work. The objective of the project was to develop a machine learning and AI-based prototype to automate the content analysis and keywording of publications, support the development of Estonian language technologies, reduce costs, increase speed and ensure the objectivity and uniformity of keywords. A prototype was developed that is capable of analysing the content of a publication and automatically generating keywords for it if the full text is available.

The Estonian Tax and Customs Board leverages AI for more efficient tax collection. The kratt for envelope wages identifies potential payers of envelope wages using a risk model.

A Python-based framework that pre-processes X-Road logs, prepares and forwards usage reports to members, displays anomalies based on moving averages, and anonymises and publishes log data as open data. The system can provide guidance to users based on anomalies and meets the requirements for open data in the public sector.

A Python-based framework that pre-processes X-Road logs, prepares and forwards usage reports to members, displays anomalies based on moving averages, and anonymises and publishes log data as open data was developed. As a result of the project, the Information System Authority now has an overview of the institutions that use X-Road and how they use it. The system can provide guidance to users based on anomalies and meets the requirements for open data in the public sector.

The kratt detects potential VAT fraud. The purpose of the project was to identify inaccurately filed VAT refund claims, whether done intentionally (fraud) or due to human error. The developed machine learning model enables scoring of future VAT refund claims, allowing the determination of the probability of whether a refund claim is correctly or incorrectly filed.

The Social Insurance Board collaborated with Feelingstream to carry out a pilot project for the detailed analysis of customer calls and online chats to identify patterns in customer enquiries. As part of the project, the Feelingstream application transcribed the calls made to the Social Insurance Board from May to August 2020 (three months) in both Estonian and Russian.

The species recognition software is designed to determine the abundance of a species using the REM method. The system allows for a project to be created for a specific survey area and duration, on the basis of which the abundance of species in the area is calculated. The system saves the images captured by trail cameras under the corresponding project, after which they are classified using AI and the REM method is applied to determine the abundance of species. The system enables the generation of reports based on the images collected under the project, following the structure defined in the terms of reference, along with the correction of observation cases with imprecise classifications and the training of the AI model using new images

The project had two main objectives: 1) to enhance the usability of the State Portal and 2) create tools for the replication of analyses. Data analysis and process mining were used to identify the bottlenecks in the portal and to propose improvements, including prototypes. In addition, tools were developed to make it easier for product owners to decide which parts of the State Portal need the most attention to enhance user experience and user feedback.

The B�rokratt personal data anonymisation application was completed in late 2022. The aim of the application is to make the training of B�rokratt enquiries even more secure. The data anonymiser is able to detect personal data name entities, such as names, personal identification codes and locations of persons, within a given text and replace them with another value of the same entity class (for example, �Tallinn� is replaced with the entity �Tartu�).
The application is designed to be used not only in B�rokratt but also in the applications and information systems of other authorities that are faced with the processing or storage of personal data.
The anonymiser helps mitigate the risk of personal data processing and complies with the regulations of the GDPR. The application enables the use of both anonymisation and pseudonymisation methods, depending on the user�s needs. Integrated clients can use the application based on the NER corpora developed by the University of Tartu and further train the corpus with institution-specific data sets.

The first automated border control gates, or ABC gates, started operating at Tallinn Airport and the Narva road border crossing point in early 2021. Automated border control gates based on biometrics speed up border crossings and provide border guards with an additional tool for identifying people and verifying their right of entry. The identification algorithms used in ABC gates minimise the risk of people crossing the border with false documents. Furthermore, preliminary tests have shown that the average time required to pass through the ABC gates and cross the border is 15 seconds.