The Tech-in-GOV call for projects

During the first stage of the Tech-in-GOV process, an initial analysis of the applications is carried out by a group of experts made up of members of the Ministry for Digitalisation, the Government IT Centre (CTIE) and the GovTech Lab. This initial analysis is based in particular on a series of predefined eligibility criteria (in French).

Applications are grouped according to the category identified by the project leaders: AI, data or interoperability. If the project leader has not specified which category their project falls into, or if the category indicated is not appropriate, the group of experts assigns the application to the category it considers most suitable.

Projects for which the solution is not clearly identifiable are taken directly by the GovTech Lab team, which examines the possibility of launching a call for solutions via its own distribution channels. These projects are then removed from the Tech-in-GOV process. The applications that are off-topic or cannot be assigned to a category, and applications for which solutions already exist or are under development, will be rejected.

In the second stage of the Tech-in-GOV process, project leaders who have successfully completed the first stage are invited to participate in a pitch session dedicated to the theme under which their project has been classified. These project leaders then can present their project in detail to an expert committee  specially constituted for each category, which can then better understand the challenges and feasibility of the projects, as well as their potential impact. Each committee together with the directors of the Ministry and the CTIE then select the projects that will move on to the third stage of the process. This narrowing down of eligible projects is also necessary to ensure adherence to the budget limits set out for the call for projects.

The project selection criteria vary depending on the category. Overall, the following elements are taken into account:  

• experimental potential (ability to push the boundaries of the state of the art in public services)
• strategic alignment with public policies (national strategies, coalition agreements)
• cross-cutting nature of the project, particularly in terms of interoperability
• innovation and contribution of new solutions that will benefit a large number of agents, citizens or users;
• complexity of the project and its ability to overcome technical challenges;
• maturity of the idea;
• speed of implementation and/or limited cost of the project;
• contribution to improving existing interoperable infrastructure;
• ethics and responsibility in the use of data and AI technologies;
• ability to produce measurable, long-term results for public services.

The degree of innovation and experimental nature are priority filters. Projects that are part of an existing programme or rely on  technologies which are already available or are currently being rolled out may be redirected towards traditional development channels.

For projects selected following the pitch sessions and for which a clear solution has been identified, the project leader works closely with the experts responsible for monitoring the project to draw up detailed specifications with a view to finding a suitable service provider. During the project implementation phase, support is provided by the team of experts in question. The GovTech Lab process is initiated for projects for which no suitable solution has been identified.

The Zenter fir d'Lëtzebuerger Sprooch (ZLS – Center for Luxembourgish Language) is proposing the establishment of a multilingual parallel corpus. This corpus will focus on Luxembourgish as the source language — and therefore the main language — with translations into three target languages: German, French and English. This multilingual parallel corpus aims to directly address the current lack of high-quality multilingual datasets focused on Luxembourgish. It will be a valuable resource for supporting advanced machine translation, training artificial intelligence language models, computational linguistics research, and other technological applications, both in Luxembourg and internationally.

The project supports the ‘Information Requests’ department of the National Commission for Data Protection (CNPD), which is faced with a growing volume of requests received by email, post and telephone.

With the gradual increase in archived questions and answers, manual searches carried out by staff had become particularly time-consuming. AI4DI automates the search for relevant information within the existing database and generates an initial draft response.

The solution is based on a computer tool incorporating a large language model (LLM), capable of effectively identifying useful content and proposing appropriate responses, while remaining within the scope of the CNPD's remit.

The "Information Requests" service of the National Commission for Data Protection (CNPD) receives a wide range of questions via various communication channels, such as email, post and telephone.

Over time, the volume of questions and answers has increased significantly, making manual searches by staff increasingly time-consuming. Through this project, the CNPD  automates both the search for information and the generation of an initial draft response, using the existing database.

The proposed solution is a computer tool based on a large language model (LLM) powered by artificial intelligence, capable of efficiently finding relevant information and providing appropriate responses.

The Luxembourg Office of Accreditation and Surveillance (OLAS) wishes to develop a tool (‘Legal and normative text analyser’) capable of performing the following tasks:

• Analyse documentation available in recognised and reliable databases such as Legilux, EUR-Lex, ISO, CEN, EA, ILAC, IAF and locally stored files;
• Identify and collect relevant documents or information demonstrating compliance with the requirements applicable to the accreditation of specific activities;
• Analyse the content of these documents or information in order to identify and extract the applicable requirements necessary for the accreditation of conformity assessment bodies;
• Prepare detailed reports based on the results of the analysis.

The ScreenreaderLB solution, developed by the  Government Information and Press Service (SIP), was launched in January as part of a soft launch. This first version offers a screen reader dedicated to Luxembourgish, meeting the need for accessibility of digital content for blind or visually impaired people.

Designed to offer speech synthesis adapted to the Luxembourgish language, the solution allows users to interact more naturally with online content. This initial launch phase will enable feedback to be gathered from early users in order to identify areas for improvement and develop the product in future versions, with the aim of gradually optimising the browsing experience and access to information.

The AI for Legacy project looks at how artificial intelligence can be used to analyse, comment on and, if needed, refactor code from legacy tools within the Ministry of the Economy. The study focuses on evaluating the capabilities of modern AI tools to revitalise these monolithic systems, which were developed over long periods of time and have become increasingly complex and difficult to maintain. The project seeks to demonstrate how AI can offer innovative solutions to improve the readability, maintainability and performance of these ageing applications, while extending their lifespan in an effective and sustainable manner.

The Government’s Information and Press Service has an enormous media library of political photos belonging to the Luxembourg State. Until now, the work on indexing these photos, carried out manually, has been fastidious and time consuming. The AI project aimed to identify the politicians on these photos using facial recognition and generate improved metadata for the media library. Setting up this facial recognition has allowed to facilitate and accelerate the provision of the photos’ metadata, which in turn resulted in better use of the media library’s content, for the benefit of the State as a whole.

This project involved using AI to extract topographical objects from aerial images, thereby making it possible to detect new buildings and roads with a view to identifying construction areas. Extracting these objects made it possible to reduce public sector workers’ workload and increase the quality and rapidity of updating the databases held by the Land Registry and Topography Administration. A further aim of the project was to allow the creation of new data and services, and object recognition.

The aim of the project proposed by the Luxembourg National Library (Bibliothèque nationale de Luxembourg - BnL) is to improve the transcription of the texts of articles by using optical character recognition (OCR). This enhancement was a prerequisite for the automated identification of "named entities" (people, places, organisations, and dates). These entities, detected using AI techniques, form the basis for a new interactive application that allows dynamic exploration of articles, thereby improving access to the archive of digitalised newspapers and journals.

This application helps anonymise court decisions for publication on the justice internet portal. JUANO is based on an artificial intelligence engine that recognises entities to be anonymised and automatically suggests categories based on the text. This application ensures the consistency of the anonymisation of court decisions and facilitates the process of making them available to lawyers and other interested parties in compliance with data protection regulations.

The CLAVIS project proposed by the Luxembourg National Archives involves submitting digitised images of seventeenth- and eighteenth-century manuscript documents to Transkribus, a platform for AI-assisted recognition of handwriting, transcription, and searches in historic documents. Automated transcription helps to make these sources, hardly exploitable until now, legible and comprehensible again for upcoming generations of researchers. ANLux also expects the project to come up with new possibilities for the automated creation of inventories or indexes of people and places.

Through this project, the State Centre for Human Resources and Organisation Management intends to improve recruitment procedures by using artificial intelligence in order to assist their agents in checking the conditions of admissibility of applications and registrations for the general aptitude test for the Luxembourg civil service.

The Ministry of Home Affairs aims to set up automated monitoring using a system for detecting aberrant data (outlier detection) in the financial data of the municipal sector.

Through this project, the Central legislative service aims to automatically label (categorise/classify) documents based on their content, with a view to coupling a chatbot with the knowledge graph of Luxembourgish legislation.

This project concerns the implementation of a data science and machine learning program to enable the National Institute of statistics and economic studies (STATEC) to produce statistics on the mobile telephone database.

This AI conversational platform needed by the Employment Agency ADEM assists state agents who receive messages from job seekers, by formulating personalised responses. This applies to emails and enquiries received via a new Chatbot service integrated into the ADEM website.

This project allows through the analysis of existing data at ADEM to develop a support model for their agents to assess which jobseekers need which measures and level of support to access employment. This diagnostic is possible by taking into consideration the characteristics and professional profile of the jobseeker.

The 'Anomaly detection' project enables the local authority association, DEA (Distribution d'Eau des Ardennes), to use machine learning to detect water leaks in the network and identify changes in a set of data with daily and seasonal differences. The aim of this project is to increase the monitoring capabilities of the DEA, who is responsible for producing and distributing water to 29 municipalities in Luxembourg.

This project concerns an artificial intelligence component to assist judges in drafting legal documents. It analyses the context of the text being drafted in real time and automatically suggests relevant legal texts and court decisions. This system significantly reduces the time spent on legal research, while improving the accuracy and quality of the references used. By facilitating immediate access to applicable legal information, it optimises the drafting process and enhances the efficiency of the judicial process.

The Land Registry and Topography Administration is responsible for managing the country's cadastral data. This data has been systematically organised for 200 years, both in graphic form (cadastral plans and sketch maps) and in tables recording changes in property rights. Information on property rights was recorded by hand in registers structured in tables from 1824 to 1972. The objective of this project is to compare different approaches to interpreting some of these registers for a given municipality, identify the most appropriate method, and then implement it within the ACT.

As part of its efforts to digitalise and improve its agility, ADEM intends to develop or improve data of high strategic value. This project improves the data needed to strengthen the organisation's capacity to match job supply and demand, thereby facilitating professional integration.

It provides for the development of two modules in the pre-processing chain for text content in job vacancies received by ADEM from Luxembourg employers and online job platforms:

• Duplicate detector: this module will eliminate duplicate job vacancies from different sources.
• Anomaly detector: this tool will identify and exclude sentences from the text of the listing that are not related to professional skills or requirements (e.g. description of the company).

The Zenter fir d'Lëtzebuerger Sprooch (ZLS – Center for Luxembourgish Language) has developed the prototype schreifmaschinn.lu, an AI-based speech-to-text tool in various languages, including Luxembourgish. The project ‘Sproocherkennung fir d'Lëtzebuergescht’ is expanding the current hypersynchronised text and audio dataset from 55 hours to 80 hours to improve the performance of schreifmaschinn.lu. The aim is to meet two requirements: diversification of speakers and situations on the one hand, and spelling accuracy on the other.

The Government Information and Press Service (SIP) is responsible for monitoring the digital accessibility of websites and applications of the Luxembourg public sector. The Observatory is an accessible website presenting a dashboard of the main indicators of digital accessibility in Luxembourg and their evolution, such as the ranking of the most accessible websites and applications, with gold, silver and bronze labels based on an accessibility score.

This project by the Ministry of Research and Higher Education aims to explore how synthetic data can be cross-referenced and then evaluate the usefulness of such data. The difficulty of this project lies in the fact that two separately synthetic data sets do not contain the same individuals and it is therefore impossible to cross-reference them on the basis of a common identifier (which would also be artificial). However, there are statistical methods that can be used to match individuals who are similar in many respects and therefore cross-reference them on the basis of their statistical characteristics. The results obtained in this project open up new possibilities for the use of synthetic data.

With the NIF#PRESERV project, the National Archives of Luxembourg aim to define a digital preservation policy at the national level and propose a concrete implementation approach based on services associated with this policy. The project responds to recommendation No. 31 of the National Interoperability Framework (NIF), which recommends the definition of a long-term digital preservation policy for the entire public sector.

The REG4Gov project of the Department of Mobility and Transport of the Ministry of Mobility and Public Works aims to design and implement a reference architecture for basic registry systems. The expected result is the definition of a systematic and standardised approach to the development of basic registry systems that comply with the relevant regulatory requirements.

Creation of a REST API web service for the Ministry of the Economy's internal management software to facilitate access to the MM-AET (establishment authorisation) database for other administrations in the exercise of their functions.