Vectr Consulting

Zeticon: Generative Search Assistant

Imke Heusschen — Mon, 17 Feb 2025 16:27:09 +0000

Een kijkje in de wereld van de Generative Search Assistant.

Een Zeticon Plugin.

Heeft u al van de nieuwe Zeticon plugin ‘Generative Search Assistant’ gehoord of heeft u hem misschien zelfs al gebruikt? Deze plugin, ontwikkeld in samenwerking met Vectr.Consulting en Zeticon, kan het zoeken naar informatie in uw bestanden aanzienlijk vereenvoudigen en versnellen! De assistent toont u in realtime waar en in welke originele bron u uw antwoord kunt terugvinden.

Generative Search Assistant

Maar hoe werkt deze Generative Search Assistant eigenlijk? Om de toenemende stroom aan informatie te kunnen verwerken wordt er – nog voordat de daadwerkelijke search plaatsvindt – gestart met het maken van een onderverdeling van alle beschikbare documenten. Deze semantisch logische onderverdelingen, ook wel ‘chunks’ genoemd, worden niet apart opgeslagen. In plaats daarvan krijgt elke chunk zijn eigen pointer zodat de gewenste informatie op een later tijdstip eenvoudig teruggevonden kan worden. Door middel van een sentence transformer model worden de tekstuele chunks omgezet in een numerieke representatie, ook wel embedding genoemd, die vervolgens wordt opgeslagen als vector in een ‘vectorstore’. Deze ‘vectorstore’ is een speciale database geoptimaliseerd voor het zoeken op basis van vectoren.

Vervolgens gaat de daadwerkelijke zoekopdracht echt beginnen. Wanneer er een vraag gesteld wordt aan de Generative Search Assistant zijn er twee manieren waarop deze op zoek gaat naar het beste antwoord op de vraag.

Keyword Search:
De zoekopdracht wordt geanalyseerd en op basis van keywords worden zo veel mogelijk relevante chunks opgehaald uit de vectorstore.
Embedding Search:
Gebruikmakend van dezelfde tools als waarmee de documenten omgezet zijn wordt de zoekopdracht omgezet in een embedding, die vervolgens wordt vergeleken met alle chunks in de vectorstore om zo de chunks op te halen die semantisch het dichtste aanleunen bij de gestelde vraag.

Vervolgens worden de resultaten gecombineerd en wordt er met behulp van een ‘reranker’ gekozen welke chunks het relevantst zijn voor de zoekopdracht. Op basis van deze chunks en de originele vraag gaat de Generative Search Assistant vervolgens met behulp van een LLM (Large Language Models) op zoek naar het antwoord op de gestelde vraag. Dit antwoord wordt op een duidelijk leesbare manier weergegeven, vergezeld door een duidelijke bronvermelding binnen de eigen documenten.

Voordelen van de Generative Search Assistant

Werken met de Generative Search Assistant biedt u verschillende voordelen die uw zoekervaring significant kunnen verbeteren:

Tijdsbesparing: U hoeft niet eindeloos te zoeken door mappen of bladeren in documenten; stel uw vraag aan de assistent en u ontvangt direct uw antwoord.
Veilige segmentatie: Doordat uw documenten via contexten worden gescheiden, krijgt u alleen relevante informatie te zien, wat de efficiëntie van de assistent verhoogt.
Menselijke interactie: Het LLM levert antwoorden op een natuurlijke, leesbare manier, zonder dat u kennis nodig hebt van complexe zoekopdrachten of query-talen.

Large Language Model (LLM)

Om te begrijpen hoe chunks worden omgezet tot een zinnig antwoord is het goed om te weten wat een LLM is en wat het precies doet. Een Large Language Model (LLM) is een vorm van Artificiële Intelligentie die getraind is op enorme hoeveelheden tekst. Hierdoor zijn deze modellen in staat om ‘almost human’ teksten te genereren, vragen te beantwoorden en snel complexe taalpatronen te begrijpen. Dit alles is mogelijk doordat LLM’s gebruik maken van deep learning om patronen te herkennen in onze taal. Deze patronen worden vervolgens gebruikt om relevante antwoorden en suggesties te genereren. Hierbij geldt, hoe groter de dataset waarop het model getraind is, hoe beter de prestaties.

Door gebruik te maken van een vooraf getraind model kan de Generative Search Assistant in realtime snelle en accurate antwoorden geven op elke vraag gesteld over de inhoud van de opgeslagen documenten.

Met deze combinatie van semantisch zoeken en AI-gestuurde antwoorden maakt de Generative Search Assistant het eenvoudiger dan ooit om snel en betrouwbaar de juiste informatie te vinden.

Wilt u weten hoe deze technologie uw werkprocessen kan verbeteren? Neem contact met ons op, dan kijken we samen hoe AI u kan ondersteunen!

Zeticon is een Belgisch software-en consultancybedrijf dat gespecialiseerd is in Intelligent Informatiemanagement en archivering.

Zeticon transformeert jouw versnipperde en slapende data in gecentraliseerde en intelligente informatie. Dankzij de Open REST-API integreert ons platform naadloos met de dominante applicaties (zoals ERP, CRM, DMS, …) binnen uw organisatie. Als pionier in Intelligent Information Management en Archiving centraliseren wij al uw bedrijfsinformatie op één veilig platform en maken het toegankelijk waar en wanneer u het nodig hebt waardoor de productiviteit toeneemt, processen en diensten verbeteren en risico’s in uw hele onderneming afnemen.

Meer informatie op zeticon.com

Large Language Model (LLM)

Met deze combinatie van semantisch zoeken en AI-gestuurde antwoorden maakt de Generative Search Assistant het eenvoudiger dan ooit om snel en betrouwbaar de juiste informatie te vinden.

Wilt u weten hoe deze technologie uw werkprocessen kan verbeteren? Neem contact met ons op, dan kijken we samen hoe AI u kan ondersteunen!

Zeticon is een Belgisch software-en consultancybedrijf dat gespecialiseerd is in Intelligent Informatiemanagement en archivering.

Meer informatie op zeticon.com

Let your data work for you. Today.

The post Zeticon: Generative Search Assistant appeared first on Vectr Consulting.

Fraud detection

Imke Heusschen — Fri, 08 Nov 2024 07:54:15 +0000

Fraud detection

Fraud detection with graph analytics

A confidential project for the Government to support public officers with the detection of fraud in a business field where the government offers subsidies. It is known that fraud occurs, and the question was whether we can use the existing data to support the officers better in the field.

Business goal : Increase the fraud detection rate.

Workflow: The data is initially analyzed using standard outlier-detection techniques in regression models. Having 11 known fraud cases within the data set allowed us to test model hypotheses. This yielded good models for most of the known fraud cases, but some remained under the radar.

We extended the research into the time and connection domains. We built time series to find behavioural patterns in time, and outliers thereof. This allowed a refinement of the detection model, but still missed two important known frauds.

The final step was to connect all data and build an interaction graph of all connected entities in the data model. This was done using a specialized graph database and apply graph analytics on it.

Outcome: These analyses actually have a broad application in business cases.

The result was astonishing; not only were the two remaining known fraud cases clearly picked out, a lot of other candidate cases showed up.
A tool for the officers, with which they can play around in a multi-dimensional domain model and they can then focus their field visits on the peculiarities from the model.

The tool we’ve created to detect fraud with service vouchers for WSE, Vlaamse overheid is clearly paying off. De Tijd noticed our remarkable results and wrote an interesting article about it.

Let your data work for you. Today.

The post Fraud detection appeared first on Vectr Consulting.

Electrolux

Imke Heusschen — Thu, 10 Oct 2024 14:39:13 +0000

Electrolux

Putting data to work for Smarter Solutions

Innovation goes beyond technology — It’s about empowering people to drive meaningful change. This principle is central to Sensr’s collaboration with Electrolux, a global leader in home appliances. Together, they’ve built a scalable cloud platform that optimizes production processes, enhances efficiency, and supports sustainability across multiple manufacturing sites.

By centralizing production data, Electrolux can detect defects faster, optimize furnace usage, and improve the accuracy of OEE reporting. Linking quality control results with production data offers valuable insights that help reduce waste, improve product quality, and lower emissions—key elements of Electrolux’s sustainability strategy.

Sensr’s solutions emphasize the critical balance between advanced digital tools and human expertise. Operators play a vital role in refining data and ensuring continuous process improvements, proving that technological innovation and human input go hand in hand in creating a more sustainable future for manufacturing.

Learn more about how Sensr is shaping the future of production here.

Electrolux is a leading global appliance company, shaping living for the better by reinventing taste, care, and well-being experiences for millions of people. With a strong commitment to sustainability and innovation, Electrolux develops kitchen, cleaning, and well-being products that meet the needs of modern living, ensuring efficiency, design, and performance.

Two key takeaways: first, this example shows that the human role remains essential in the synergy between technology and human expertise. It’s the employees of Electrolux who, together with advanced digital tools, ensure continuous improvement of the production process. Secondly, these applications help Electrolux remain not only technologically innovative but also human-centered and responsible—another key pillar in their sustainability framework.

This kind of success is only possible through strong collaboration. Working with Cronos aan de Leie and Electrolux has been key to drive both innovation and efficiency. At Sensr, we’re proud of what we’ve achieved so far and look forward to seeing where this partnership will go, as we continue combining technology with human expertise for future improvements.

More info on electrolux.be

Let your data work for you. Today.

The post Electrolux appeared first on Vectr Consulting.

Departement of Work and Social Economy

Imke Heusschen — Wed, 09 Oct 2024 14:57:38 +0000

Departement of Work and Social Economy

A growing support with Data & AI.

Within this department, the Data division oversees a range of data-related activities. A key component is the Data Science and AI team, which focuses on innovative use cases that enhance the value of the Inspection and operational departments. This involves leveraging data research, advanced data science techniques, and AI engineering.
What benefits do these innovations bring to Ex Ante (preemptive) and Ex Post (retrospective) inspections? What impactful use cases can be created for the operational departments? And how have Vectr.Group consultants played a role in this?

Vectr.Consulting explored these questions with Emmanuel Thienpont, Product Manager for Data Science and Artificial Intelligence, and Roeland Juchtmans, Project Manager for the WSE Data division.

In 2016, the initial steps were taken to explore innovative approaches using data and data science methods. By late 2017, as various use cases gained traction, the need for additional expertise arose. This led to a promising collaboration with Vectr.Consulting, providing the necessary support. Around the same time, WSE made a pivotal decision to adopt a Cloud provider, with the Data Science and AI team playing a key role. This move was essential, as a Cloud environment is critical for the successful development of Data and AI use cases.

Project-specific platforms and technologies were chosen as needed.

Team Prometheus –
Greek myth or Science fiction?

A significant achievement in 2018 was the development of a fraud detection system for service vouchers, which continues to successfully support the Flemish Social Inspection. More information can be found here. This confirmed the immense value of evidence-based data research for WSE. The success of this tool immediately sparked new ideas and use cases, such as budget forecasts for service vouchers, process flow analysis through process mining, and research into parliamentary inquiries. Recently, the rise of Generative AI has opened even more possibilities.

In early 2024, the team welcomed Roeland, whose experience with the Social Economy department and expertise in data analysis brought fresh insights into what truly adds value for WSE’s operational services.
Around this time, the team was renamed Team Prometheus, symbolizing the future and the potential they see in combining existing data with modern technologies. Neither myth nor science fiction, Team Prometheus strives for practical and impactful use cases within their domain.
Over the years, numerous projects and test cases have been developed for the business, with increasing success and proven value for both the Inspection department and operational departments. These two focus areas, represented by a slight division within the team, reflect the variety of data and audiences they serve.

Evidence-based data inspectors

As mentioned earlier, the fraud detection system has proven highly valuable for the Flemish Social Inspection within WSE, inspiring the department to seek more data-driven support.

Previously, inspectors handled their own data analysis during investigations, but this approach began to hit its limits as the complexity and volume of data from increasingly large companies and organizations grew.

Since 2023, Team Prometheus has been assisting inspectors in larger cases, such as service vouchers and labor and professional cards, by delivering comprehensive analyses from multiple angles. They integrate complex datasets from various sources, including internal, government, judicial, and inspection-specific data, to uncover insights and connections between data points. By deeply analyzing and enriching the data, these aggregates provide inspectors with additional evidence and validation for their investigations. This close collaboration positions the team as “data inspectors” working alongside actual inspectors throughout the investigative process.

The results are compelling, with major cases regularly making media headlines. In the future, this type of investigation is expected to expand across various domains, with modern technologies like generative AI and graph analytics offering significant added value. Several use cases are already being explored, such as providing data-driven recommendations or alerts to case managers for recognition applications.

“The experts from Vectr are not worth their weight in gold;
they are gems. Each one is a reservoir of knowledge, further strengthened by their colleagues within the Vectr group.”

“The experts from Vectr are not worth their weight in gold; they are gems. Each one is a reservoir of knowledge, further strengthened by their colleagues within the Vectr group.”

Emmanuel Thienport, Product Manager Data Science & Artificial Intelligence

A positive contribution to operational services

As in many other organizations, operational employees are continuously occupied with planning, completing tasks, and responding to inquiries from customers and colleagues. While they are already data-driven and rely on their deep domain expertise, the fast-paced environment can make it difficult to think innovatively, particularly when it comes to technological advancements outside their usual focus. To overcome this challenge, Team Prometheus actively collaborates with the departments, targeting areas where they can deliver meaningful added value.

Frequently given answers

A highly promising use case with significant potential lies in the realm of Frequently Asked Questions (FAQ). With the growing availability of NLP technologies and new possibilities brought by generative AI, textual data can now be analyzed more efficiently and accurately.
At WSE, a large volume of unstructured data—such as the numerous daily emails—makes it nearly impossible to manually process and analyze everything, causing valuable insights to be overlooked.

One initial test case, developed in collaboration with the Department Organization Management and Development, focuses on a segment of this data: email inquiries related to training vouchers. The questions are categorized into themes, and a top five of the most frequently asked questions per theme is generated, along with the corresponding answers provided by the department. This process leverages data engineering, data science, NLP techniques, and generative AI. The outcome serves as a foundation for an FAQ truly based on actual questions and commonly given answers. If an FAQ already exists, this approach can help assess its quality and update it using rapidly available information.

Another compelling case involves the 1700 helpline, a frontline customer service that handles numerous inquiries—some structured through a scenario menu—while others, such as case status requests, are not answered directly. These are forwarded to various departments, including WSE. Annually, thousands of questions, such as those regarding encouragement bonuses, flood the operational service. By categorizing and filtering the most common queries, internal processes can be streamlined, and the information provided improved. This helps reduce the volume of incoming inquiries or enhances the 1700 line’s script. Once again, the combination of generative AI techniques with data science methods like clustering and classification has proven to be a key to success. The team collaborates closely with the Secretary General’s departments to iteratively enhance this knowledge base.

What does the future hold?

The increasing volume and availability of data present expanding opportunities for discovering connections and gaining deeper insights in the future. One potential application is calculating risk scores using available data, integrating findings and anomalies from past analyses that surpass what a human analyst could achieve. This will be accomplished through a combination of data science techniques, machine learning engineering, and domain expertise. Besides providing valuable insights, this approach will significantly enhance efficiency by automating a substantial portion of the research work, allowing
employees to focus more on cases requiring additional attention.

Additionally, the use of generative AI enhances the accessibility and comprehensibility of the vast amounts of textual data. This opens up opportunities to implement both established use cases and explore new applications in this field. For many, including those at WSE, this represents a new frontier, which can make it challenging to gauge the potential or overcome resistance to changing traditional methods. By starting with smaller-scale projects, a gradual shift in perspective can be encouraged, provided these projects deliver genuine value to the departments involved. Without clear added value, departments may be reluctant to dedicate their time and resources to validating and testing new solutions, an engagement that is essential for success.

Generatieve AI one-size-fits-all?

The answer is straightforward: No, Generative AI (GenAI) is not the solution for everything. While working with large volumes of textual data is new for many, the term “Generative AI” is often quickly associated with these scenarios. In reality, however, the real value lies in integrating traditional data science methods with existing data environments, where AI or GenAI is applied to specific aspects of the process. It therefore requires in-depth knowledge of different data domains. Expertise in GenAI alone is far from sufficient.

This perspective aligns with the notion that GenAI should not be the starting point of a project but rather a tool to enhance a solution. For instance, using MS Azure OpenAI as a GenAI application to clarify core questions from often ambiguous initial queries exemplifies this approach. While Natural Language Processing (NLP) can clean up language by removing stopwords, it cannot rewrite the text or the question itself. GenAI, on the other hand, offers significant benefits by restructuring or rewriting source data (text), which improves the efficiency of data (the questions) categorization. Consequently, vague questions can be organized more effectively, allowing a Large Language Model (LLM) to provide relevant content within the correct context.

“The consultants at Vectr.Consulting
are highly motivated and supportive, excel in teamwork,
and contribute valuable insights.”

Roeland Juchtmans, Product Manager Data

AI strategy meets bottom up data innovation

Team Prometheus works closely with colleagues responsible for delivering services and conducting inspections. This collaboration is increasing awareness of data science and AI possibilities among more team members, who are keen to leverage these technologies. For successful outcomes, senior management’s support and active promotion of this process are crucial. This commitment is evident from both the current and former Secretary-Generals, as well as from the management’s dedication to a robust AI strategy. Consequently, a “coalition of the willing” has formed, and it is hoped that this will continue to expand. Support from the workforce, combined with the strategic vision from senior management, is essential to making these applications operational.

The Department of Work and Social Economy is responsible for employment and social economy policies. This includes policy preparation, follow-up, monitoring, and inspection, along with managing the European Social Fund (via Europe WSE) in Flanders.

More information on vlaanderen.be

Sharing is caring –The data science working group

Founded by Emmanuel Thienpont in 2017, the Data Science Working Group within the Flemish Government aims to foster collaboration on innovative data initiatives. This initiative led to a partnership with MOW (Mobility and Public Works) to address parliamentary questions, building on a proof of concept (POC) previously developed within WSE at the request of the Secretary-General. Vectr.Consulting’s consultants played a key role in supporting both departments to achieve this project.

The strength of Partnership

Vectr.Consuling is committed to establishing enduring, high-quality relationships with its clients. Building trust, collaborating effectively, and recognizing each other’s strengths and weaknesses are key. Delivering the right expertise at the right time is vital.

Sharing is caring –The data science working group

The strength of Partnership

With the support of Cronos Public Services, Vectr.Consuling is committed to establishing enduring, high-quality relationships with its clients. Building trust, collaborating effectively, and recognizing each other’s strengths and weaknesses are key. Delivering the right expertise at the right time is vital.

More information on vlaanderen.be

Let your data work for you. Today.

The post Departement of Work and Social Economy appeared first on Vectr Consulting.

MOW – Detecting Water Bodies

Imke Heusschen — Fri, 05 Apr 2024 12:56:30 +0000

MOW – Detecting Water Bodies

A more efficient view on water

The ports in our country serve as vital economic hubs and demonstrate a commitment to ecological awareness. The development of a port reflects our relationship with the environment and our methods of separating water from land to ensure the controlled management of port activities. This segregation is achieved through various means, including the implementation of seawall techniques (also referenced in our case study on HARO blocks). While these techniques are relatively straightforward to chart due to their fixed structure and location, mapping out the movements of water bodies crashing against these structures poses a greater challenge. This challenge has been a focal point for the Department of Mobility and Public Works (MOW) of the Flemish Government for some time. This is where Vectr.Consulting steps in to provide assistance.

Automatic labeling

The General Technical Support Department (ATO) assists the Department of Mobility and Public Works in mapping images around bodies of water. Using drones, the ATO cell captures a multitude of aerial photographs of port infrastructure through photogrammetry. Optical cameras create 3D models, within which delineated objects such as locks or breakwaters are relatively easy to distinguish. However, 3D rendering software struggles with bodies of water. Currently, many of these bodies of water are manually masked. To streamline this time-consuming process, we are collaborating with MOW to develop a machine-learning model that is automatically applied to all new drone photos.

Based on the manual labeling already conducted, we train our own model, called the Segmantic Segmentation Model. Simply put, in a drone photo of a body of water, each pixel where water is detected is marked as a 0, and where it is not detected, as a 1. Metrics calculations, such as Intersection over Union (IoU), are used in ML processes to measure accuracy and error margins of this 0/1 localization. The results are displayed in a score. MOW’s trained model achieves a score of 0.985 in test jobs, meaning only a very small number of pixels are incorrectly labeled.

The client can upload all images it captures to an Azure Storage Container, where they enter a data pipeline to be analyzed by the model. Subsequently, the output images are placed in a second Storage Container where the client can download them. Client feedback allows us to adjust and optimize the model accordingly. The model is written in Python with Pytorch as the Deep Learning framework. Azure Machine Learning, a well-known cloud service from Microsoft, is used to manage our training and test jobs. Additionally, the model is tracked in the AzureML Model registry to manage all trained models. Finally, all infrastructure is managed using Terraform, an infrastructure-as-code tool.

A succesful continuation

As experts in data intelligence, we are extremely proud to assist MOW in the realm of machine learning with object detection and ML ops. This collaboration was made possible through Cronos Public Services, a division of the Cronos Group that aggregates and facilitates the expertise of its subsidiary companies for government services. For quite some time now, Cronos has maintained a framework contract with the department. With the support of Cronos and MOW, Vectr.Consulting can continue its efforts to enhance the effectiveness and safety of our surroundings by deploying appropriate technology.

Every time there have been new drone flights around the port, our AI application is used to mark bodies of water. This ultimately results in serious time savings for drone pilots and more accurate labeling of the water bodies. Work is also ongoing to improve the model to perform optimally in other areas, such as rivers and canals.

In addition to detecting bodies of water, we also assist MOW with other ML projects. Be sure to check out our case studies on detecting HARO block movements and Unsafe traffic situations.

The Mobility and Public Works Department (MOW) of the Flemish government makes Flanders move: by road, by water and by air. Their experts on the move work every day for safe, smart and sustainable mobility and for today’s road and (air)port infrastructure. The MOW Department supports the Minister’s policy in terms of content, both for mobility and traffic safety and for investment, management and operation of the transport and port infrastructure. The MOW Department also performs administrative and technical support tasks including expertise and advice on concrete and steel structures, ground tests, dynamic traffic management, hydrological research, and so on. Finally, the Department also guarantees maritime access to the Flemish seaports.

More info on vlaanderen.be

A succesful continuation

In addition to detecting bodies of water, we also assist MOW with other ML projects. Be sure to check out our case studies on detecting HARO block movements and Unsafe traffic situations.

More info on vlaanderen.be

Let your data work for you. Today.

The post MOW – Detecting Water Bodies appeared first on Vectr Consulting.

MOW – HARO Blocks

Imke Heusschen — Fri, 16 Feb 2024 06:48:19 +0000

MOW – HARO Blocks

Stability monitoring of sea breakers through AI-driven object detection.

Anyone who has sailed into the ports of Ostend or Zeebrugge is undoubtedly familiar with the concrete blocks that protect the sea breakers. These blocks, called HARO blocks, are intended to shield the lower-lying polder areas from strong waves. The sea is in constant battle with the land to conquer territory. In response to the sea’s actions, there is essentially a choice between hard coastal defense techniques and soft coast protections.

When it comes to robust solutions, such as building breakwaters, groynes, dikes, and the familiar concrete blocks, they all face the powerful force of waves, leading to shifts and movements. To track these shifts, Vectr.Consulting supports the Department of Mobility and Public Works (MOW) of the Flemish Government, responsible for the management of these blocks, develop a shift and movement detection tool.

To measure is to know

Every year, we capture drone footage of the HARO blocks, which is later processed into an orthophoto – a digitally enhanced aerial image with superior geometric accuracy compared to regular aerial photos. Thanks to their distinct shape, the blocks and their orientation are quickly and accurately detectable through a custom-trained object-detection model. Using object detection algorithms, we identify the HARO blocks as vectors in the three-dimensional space.
For each identified HARO block, we pinpoint the center of the bounding box – a virtual box that encompasses the entire block – as a pixel coordinate. These pixel coordinates are then converted to Lambert72 coordinates, a geometry projection type used in Belgium for creating highly precise topographic maps, leveraging the characteristics of the orthophoto.
The objective is to detect all HARO blocks in these orthophotos and assign them a vector in the Belgian coordinate system. This process facilitates the measurement of shifts and tilts over the years.

The tool is coded in Python, and we preprocess the images using OpenCV, an open-source library equipped with functions for real-time image recognition. Afterwards, we label the images along with their corresponding data. The object detection model is trained using the Azure Custom Vision API. From there, the results are processed as coordinate vectors.
We handle all infrastructure using Terraform, an infrastructure-as-code tool that allows us to manage the data infrastructure effectively.

Collaboration for smarter prevention

As data intelligence experts, we are proud to support MOW in the realm of machine learning, specifically in object detection and ML ops. This collaboration has been facilitated through Cronos Public Services, a sector within the Cronos Group that consolidates and leverages the expertise of its subsidiary companies for government services. Cronos Public Services has maintained a long-standing framework contract with the department.

Currently, MOW is in the process of exploring the solution using historical photos, with future plans to seamlessly integrate it into their existing application. This enhanced application will imbue the new orthophotos with the necessary intelligence. The integration is designed to improve their ability to predict maintenance needs for these blocks, whether it involves replacement or repositioning. Consequently, coastal areas around Ostend and Zeebrugge will be safeguarded from wave impact and erosion, concurrently bolstering the economic position of the ports. It represents a more intelligent collaboration with nature.

In addition to detecting HARO block movements, we also assist MOW with other ML projects. Be sure to check out our case studies on detecting unsafe traffic situations and water bodies.

More info on vlaanderen.be

Collaboration for smarter prevention

In addition to detecting HARO block movements, we also assist MOW with other ML projects. Be sure to check out our case studies on detecting unsafe traffic situations and water bodies.

More info on vlaanderen.be

Let your data work for you. Today.

The post MOW – HARO Blocks appeared first on Vectr Consulting.

VDAB

Imke Heusschen — Fri, 12 Jan 2024 14:29:35 +0000

VDAB

Applying AI for a smart job market.

The job market in Flanders is a complex and dynamic phenomenon. Not everyone finds their way to the right job easily. There is a tremendous opportunity to make a guided job search smarter. What skills and competences has the candidate gained in past experiences? How do these align with a better job? What information can be derived implicitly or explicitly from the candidate’s search behavior? But also, how do we guide an active worker to a job that is a better fit than the current one? Many questions that VDAB is seeking the appropriate answers to, can increasingly be supported by artificial intelligence (AI). Technology that swiftly and efficiently guides citizens to the right job is more relevant today than ever before.

An efficient ecosystem solution

Thanks to a driving force from management towards the application of innovative technologies, VDAB started in 2016 with the installment of an innovation team that, at the time, had the freedom to explore where innovative technologies like AI and machine learning could assist VDAB. The team also got the task of centralizing all associated technical knowledge – which had a strong tendency to fragment across the organization – for the benefit of more effective functioning for both itself and the citizen. To reinforce this knowledge, the innovation team, led by Karolien Scheerlinck, looked at collaborating with various external experts in their domain, providing the vital cross-pollination that the project needed. Today, the AI team consists of 32 individuals and external experts, intentionally accounting for over 30% of the team.

A set of intelligent tools

VDAB aims to harness the power of technology across various projects that support its policies and together form an integrated solution for an improved job-search experience for citizens. As Karolien emphasizes, there are many more possibilities to ascertain what a candidate-employee wants to do with his career and, in which directions he wants to move. VDAB wants to gain insights into the job careers and technology plays a key role in this. Vectr.Consulting helps VDAB implementing innovative technologies in a variety of business projects, summarized below.

Jobbereik

Karolien: “For many people, the job market is complex and dynamic. They often have a limited understanding of their skills and competences and the professions they can pursue with them. Even for employees, it is not always clear which skills they should specifically look for. Jobbereik provides insight into this.“

Jobbereik is an orientation tool that maps all the skills of the candidate-employee based on his experience and education on to the skills required for a job. It builds a Knowledge Graph, a network of semantic metadata, that links job seekers, job vacancies, and training programs to each other through skills. These links form as they naturally occur in reality: citizens possess skills that are required for job vacancies, which, in turn, belong to types of professions that are in demand by citizens. The skill sets are defined per job occupation, based on data from the job market that is always up to date.
The tool uses Python as application technology offering end users to send queries, and the semantic network itself is stored in a Neo4J database. The application is released on various platforms via Docker containers. The tool itself is already active, and feedback from over 180,000 active users indicates that 80% of users are satisfied with the tool.

Jobmap

Jobmap visualizes the job landscape within Flanders. It illustrates how professions relate to each other within the job market.  Within VDAB, the Competent ontology is employed. This is a standardized description of occupations and their corresponding competencies, serving as a common language in the labor market. The drawback of such a standard is that it lags a bit behind the latest changes in the job market. Through Jobmap, we provide a data-driven application that helps keep the Competent standard up to date with the latest developments in the labor market.
For example, the tool offers insights into clusters of jobs vacancies where no standardized profession exists. If a person searches for terms like data, development, and intelligence, the result is often a mix of professions, while that cluster actually represents the skills of a specific profession such as a data scientist. From these insights, it becomes easier to identify and define new, emerging professions.
The solution also assists job seekers by offering them a visual search for vacancies. For instance, when a teacher enters the term “Pedagogy”, he can visually see which potential job vacancies align with this search term. You can think of it as an eye- opener in your search for suitable employment.

Jobmap is based on a t-SNE reduction technique. This is a machine learning algorithm suitable for converting the complex accuracy of data – or embedding it – into coordinates of lower dimensions (2D or 3D), which are easier to interpret for a user. The application is written in Angular with regl-scatterplot/webgl, a visualization method to illustrate correlations between points on a graph.

Skill Navigator

Skills are key to employment and prosperity. With the right skills, people are prepared for
high-quality jobs. For businesses, a proper identification of skills for work is crucial. They must compete globally, innovate, and quickly adapt to new trends. To maintain their position as leaders in innovation, they need to continue delivering the right people with the right skills.

Skill Navigator is an assisting tool that makes skills searchable and visualizes similarities. It allows citizens to input any task in their own language (e.g.,’I can knead dough’) and link it to skills, competencies, and professions. The tool is used within Jobbereik to search for similar skills within the skill profile. Moreover, it also takes into account 21st-century skills such as safety and teamwork, which are increasingly required in today’s job market.
Underlying the tool is a model based on a universal sentence encoder, which converts text into a vector. This vector is then matched through an OpenSearch database to precalculated encodings, where categorical variables are converted into numeric values to be more easily interpreted by a ML model.

“The collaboration between VDAB

en Vectr.Consulting has always been very smooth.”

Karolien Scheerlinck

The power of cross-pollination

The collaboration between VDAB and Vectr.Consulting started four years ago under the leadership of Cronos Public Services, the subdomain of the Cronos Group that bundles and facilitates the expertise of its subsidiary companies for government services. As an expert in data intelligence, Vectr.Consulting assists VDAB in the fields of data science, artificial intelligence and machine-learning operations. Both parties understand each other very well and always seek pragmatic solutions. Solutions that not only revolve around the technology itself but also focus on the people within the project. If the employee or VDAB feels that there is no match at the project level, the paths separate for the benefit of both parties. Building such a future vision regarding work only makes sense if the team behind it also has a good understanding and complements each other well.

This makes VDAB an excellent environment to build an ecosystem of benevolent parties, wherein which knowledge flows freely. VDAB wants to shapes the ecosystem itself and brings together the various involved parties.

The Flemish Service for Employment and Vocational Training is the employment service of Flanders. It is mandated by the government to promote the better functioning of the Flemish labor market by providing job seekers and employees the opportunity for a meaningful and sustainable career. To achieve this, the service collaborates with other service providers in the sector. In addition to developing career perspectives and ownership, VDAB also offers every employer the perspective on talent and support. It strongly emphasizes competence development, career services, and facilitating labor market operations. The focal point of its competence is consolidated on the website, a job site with thousands of jobs, training
opportunities, and tools.

More info on vdab.be

The power of cross-pollination

More info on vdab.be

Let your data work for you. Today.

The post VDAB appeared first on Vectr Consulting.

MOW – Unsafe traffic situations

Imke Heusschen — Tue, 21 Nov 2023 14:22:18 +0000

MOW – Unsafe traffic situations

Prevention, better than curing, thanks to technology.

It’s better to prevent an accident than to cure it. Logical, right? Neverteless, there are dozens of traffic victims in our country every day . Whether slightly, severely or fatally injured, every accident is one too many. Accident prevention is key. This is one of the authorities of the Flemish Government’s Department of Mobility and Public Works (MOW). Recently, it introduced a new governing approach, called the Mobility’s Innovative Approach (MIA). The focus of this approach is on optimizing and streamlining communication channels between the different stakeholders and experimenting with proactive detection of unsafe traffic situations, aiming at preventing accidents. Vectr.Consulting particularly helps with the smart image recognition.

MOW – Unsafe traffic situations

Prevention, better than curing, thanks to technology.

Technology is watching over humans

Since March 2021, the department MOW has launched multiple pilot projects in Flanders whereby fixed and drone cameras detect unsafe traffic situations. In West Flanders, there are projects set up in Kortrijk, Roeselare and Zonnebeke. Here, the MOW’s data lab investigates how unsafe traffic situations involving cyclists are detected before an accident would actually occur. Drones and cameras register traffic flows (mainly at busy intersections) and detect problems with lighting, inappropriate speeds, disregarded right-of-way rules, crossability or accessibility of public spaces. For example, they can detect how a cyclist on a roundabout in Roeselare just barely avoided being hit. This does not necessarily mean that expensive infrastructural interventions are required. Sometimes the distracted behavior of the traffic user is the simple cause, which can be easily enforced.

Over a period of 4 months, we received thousands of images for analysis, which we then used to scanned for pedestrians, cyclists, buses and cars and more specifically their respective movements. To make effective evaluations of a safer solution in traffic, we initially measured calibration points on the road, to transform the pixel coordinates to geographical coordinates. With this transformation, we project the trajectory of each road user on a map.

We then analyze the observed projected trajectories. The analyses is varied, ranging from measuring traffic violations to the frequency of violations over time. More importantly, however, is to research the proactive detection of dangerous traffic intersections. Detecting near misses identifies the possibility of a dangerous intersection. Although the definition of a near miss is difficult to define, we are able to identify many conflict situations this way.

Through a secure connection, the images are transmitted to an Azure Storage Container, a repository that allows users to store and manage large amounts of unstructured data in the cloud. The application itself is written in Python and relies on a dataset of labeled data. Since we did not have access to such a labeled dataset ourselves, we used a pre-trained YOLOv5 model in Pytorch, an object detection model trained on similar camera images in Montreal (Canada). To track the many moving objects in a video recording, we rely on BYTEtrack, a state-of-the-art Multi Object Tracking algorithm that determines the identity, location and trajectory of each moving object, even in cases where the object is partially or completely covered (occluded). Processing of all camera images is done on dockers in Azure Cloud.

The trajectories are stored in Citus, a distributed PostgreSQL database, and all the infrastructure is managed with Terraform, an infrastucture-as-code tool that allows us to manage the data infrastructure.

Working together for safer traffic

As experts in data intelligence, we are very proud to help MOW in the field of machine learning with object detection and ML ops. The project came about thanks to Cronos Public Services, the sub-domain of the Cronos Group which pools and facilitates the expertise of its sub-companies for public services, and which has long maintained a framework contract with the department.

The project has since been transferred from the data lab of the Department of MOW to the Agency for Roads and Traffic (AWV), which in turn will continue to use the cameras in analyzing high-risk locations all over Flanders. The cameras will thus become a very important tool in the fight against traffic accidents. As Vectr.Consulting, we hope to do our bit to make our environment safer by deploying the right technology in the right place.

In addition to detecting unsafe traffic situations, we also assist MOW with other ML projects. Be sure to check out our case studies on detecting HARO block movements and water bodies.

More info on vlaanderen.be

Working together for safer traffic

In addition to detecting unsafe traffic situations, we also assist MOW with other ML projects. Be sure to check out our case studies on detecting HARO block movements and water bodies.

More info on vlaanderen.be

Let your data work for you. Today.

The post MOW – Unsafe traffic situations appeared first on Vectr Consulting.

LCM – The intelligent scanner

Imke Heusschen — Tue, 24 Oct 2023 07:25:38 +0000

LCM – The intelligent scanner

Even in 2023, the Christian Mutuality (LCM) receives about 40,000 mostly handwritten doctor’s certificates per day. These certificates contain information regarding the patient, the reason(s) for the doctor’s visit, and the amount to be paid for the doctor’s consult. Part of the amount is refunded by the LCM. The data of the doctor’s certificates must be digitized before they can be processed. For handwritten certificates, this requires manual labour to read from the certificate and input the data in the applications and databases. Automating the process of reading and transcribing the handwritten text reduces labour cost and speeds up the processing of the certificates.

LCM – The intelligent scanner

An intelligent solution

The solution is based on artificial intelligence (AI), augmented with advanced image processing. It is capable of automatically reading and transcribing close to 20% of all A-type doctor’s certificates. This results in a serious manual labour reduction and a significant shortening of the processing time, without loss of quality. Speedier refunds thus result in happier customers.

The AI is tuned in such a way that it does not make more mistakes than humans do, and it is monitored over time to ensure its quality does not deteriorate. Every now and then the AI needs to retrain to maintain its high standard and to handle previously unseen data.

“Thanks to the handwriting AI, we are now capable of automatically processing nearly 20% of all A-type doctor’s certificates, which greatly enhances and speeds up our patient reimbursement process.”

Roel De Spiegeleer, ICT director LCM

2-step approach

Reading handwriting on forms requires a two-step approach:

1. Finding the text

Finding the text uses advanced image processing techniques. Doctors don’t tend to write between the lines, and they don’t use the same writer’s style. Therefore, a set of algorithms is used to detect text boundaries and to homogenize the text’s look and feel.

The image-processing algorithms are also required to extract a good training set for the AI models. Examples of such algorithms are:

Fetching the boundary box of a region of interest (ROI)
Detecting the number of text lines within the ROI
The splitting of text into words
Creating uniform pen thickness of the text
Suppression of background noise
Reduction of colours
Matching against templates
Alignments

2. Reading the text

Once found on the image, the reading of the text is performed by custom-trained AI, based on neural networks and optimized for handwriting. The architecture of the neural network combines recurrent and convolutional layers.
The custom-trained AI performs much better than available third-party generalized models, because they are trained for the specific business case and know the business jargon very well.
Both the image processing, which is CPU-intensive, and the transcribing of the handwriting, which is RAM-intensive, are offered as configurable, stateless micro-services accessible through a RESTfull API. They are integrated with a front-end and back-end application, and orchestration through an external workflow engine.

The production system is run completely in the local data centre. The AI model trainings are performed using Vertex AI on the Google Cloud Platform.

Applied Technologies

The microservices are written in Python and run using Flask servers within Docker containers in an OpenShift (Kubernetes) environment. This ensures horizontal scaling and optimized tuning for either the image processing towards CPU, or transcriptions towards RAM.
The image processing algorithms are built on standard Python libraries, such as OpenCV. The AI is built using TensorFlow and is trained using Vertex AI on the Google Cloud Platform.
The AI monitoring and quality assurance is executed on a Looker Studio dashboard on the Google Cloud Platform.
Additional technologies are used for small dedicated tasks that support the AI solution. These include:

Tesseract OCR for the recognition of non-handwritten text;
Google Vision API for the recognition of handwritten text for which no custom model can be trained and for which no self-learning AI can be built;
YOLO for finding non-text objects on a form, such as stamps or signatures.

All code is integrated with the in-house software release management based on GitLab, Jenkins and Nexus.

The “Landsbond der Christelijke Mutualiteiten” (LCM), or Christian Mutuality (CM) for short, is a Belgian health insurance fund that groups nineteen regional health insurance funds. With its four and a half million members, CM is the largest health insurance fund in the country. Its mission is “keeping healthcare affordable for everyone”.

More info on cm.be

A team within a team

LCM maintains a framework contract with Cronos Public Services, the sub-domain of the Cronos Group that bundles and facilitates the expertise of its sub-companies for public services. Thanks to Cronos, the Vectr.Consulting team was integrated in the LCM development teams with the task of introducing everything AI-related that was needed for reading the doctor’s handwriting, including the continuous AI monitoring and learning capabilities. This was the largest project within the LCM to integrate such an AI within its back-end infrastructure.

A team within a team

More info on cm.be

Let your data work for you. Today.

The post LCM – The intelligent scanner appeared first on Vectr Consulting.