Deep learning requires a lot of computing power. We use Amazon Web Services to provide that power in the most flexible way.
Davio Larnout Cofounder and Business Lead, Radix.ai
  • Challenge

    VDAB, a Belgian public employment service, wanted to know how they could use machine learning to create value from their data. They wanted to translate that value in a way that would impact their strategic goals.

  • Solution

    With the flexible, scalable compute power of Amazon EC2, Radix.ai created a deep learning model to help VDAB enhance its job-matching function to improve Belgium’s labor market and better connect job seekers with jobs.

  • Benefits

    Radix.ai’s deep learning model improved job matching quality. Rather than relying on one-to-one matching, search results are now more targeted. Based on word relationships and the interests and behavior of the users, job matches align more closely to the aptitudes, talents, and preferences of job seekers.

VDAB, a public employment service (PES), helps residents of Flanders (a region in Belgium) find jobs or take vocational training.

The Radix.ai team connected with VDAB through their Innovation Lab, an internal team focused on innovative technologies and ideas. The team uses proof-of-concept (POC) projects to validate which innovations support the overall strategy of VDAB.

“The Innovation Lab team approached us with two questions,” explains Davio Larnout, cofounder and business lead at Radix.ai. “First, they asked how they could use machine learning to create value from their data. Then they wondered how to translate that in a way that would impact the strategic goals of the PES. We realized the data contained in their CVs and job postings would make a great starting point to help them achieve their mission of connecting people with jobs.”

“A core function of VDAB is matching job seekers to vacancies. We’re always looking for better ways to make matches for our users,” says Michael De Blauwe, project manager at the VDAB Innovation Lab.

“VDAB’s original job-matching tool is rules-based, which requires one-to-one matches for a successful search,” says Larnout. That means if a posting states a requirement, such as a driver’s license or specific software experience, then a match occurs only if the job seeker’s CV contains the words and phrases used in the job description. Rules-based models find it difficult to process natural language, whereas a deep learning solution captures nuance and accounts for relationships between certain phrases or user preferences. "The hosting service for the BNA website was becoming increasingly problematic," says Huda Ahmed Mohsen, chief of IT for MIA. Problems with the site's availability, performance, and security were becoming more frequent even as BNA was working to implement a much-needed redesign of both the website and its back-end publishing workflows.

Radix.ai applied machine learning to VDAB data to provide better-targeted matches for its users. Deep learning, a subset of machine learning, enables machines to mimic human behavior. “Deep learning requires a lot of computing power,” says Larnout. “We use Amazon Web Services to provide that power in the most flexible way.” Radix.ai, an AWS Partner Network (APN) Standard Consulting Partner, relies on Amazon Elastic Compute Cloud (Amazon EC2) to generate compute capacity in the cloud whenever its clients need it. The company uses Amazon CloudWatch to gain a unified view of its AWS resources and services.

There was a sense of urgency with the VDAB project, which was internally named JobNet. Radix.ai and the Innovation Lab team shared a prototype of the service with VDAB’s director. She was impressed with the results and asked that the technology be put in production within six months.

“Deploying our technology required access to on-demand computing power and flexibility to spin up different machines and services. These abilities allow regular experimentation and learning by the algorithm. The most convenient way to ensure that was through the AWS Cloud,” says Larnout.

To train Radix.ai’s deep learning model, VDAB regularly uploads new vacancies and CVs to the engine via Amazon Simple Storage Service (Amazon S3), which acts as a bridge with clients, where they can drop data. VDAB automatically loads content in an Amazon S3 bucket and Radix.ai picks it up to process it further using Amazon Simple Queue Service (Amazon SQS). With each new data set, the engine learns how the job market evolves, noting changes in job demand and how trends shift over time.

The solution also learns how jobs are spoken of and what the changing interplay of words means. For example, the job of a data scientist is relatively new. Related to that role are the jobs of machine learning engineer, data analyst, and even AI architect. “The system we created learns the meaning of those words, those new titles, from the data alone. No one has to manually feed in that information,” says Larnout.

Radix.ai’s deep learning model continues to improve matching quality that surpasses the rules-based model. Rather than relying on one-to-one matching, search results are now more targeted. Based on word relationships and the interests and behavior of the users, job matches align more closely to the aptitudes, talents, and preferences of the job seekers.

“The data from our systems reveals these relationships, creating a matching solution that is completely data-driven, which is one of the goals of the Innovation Lab,” says Eric Klewais, analytics manager for VDAB Innovation Lab.

Another benefit of Radix.ai’s deep learning model is its ability to operate in multiple languages. Inhabitants of Flanders primarily speak Dutch, but Belgium is a multilingual country that counts Dutch, French, and German as official languages. In addition, many residents and their employers speak English. This makes the technology appealing to other European public employment services that serve multilingual job seekers and employers.

The success of the JobNet launch has prompted more use cases that VDAB and Radix.ai want to explore. For example, an employer may receive notice from an employee that she plans to leave her job. This employee performed well in her job and the employer would like to replace her with someone with similar skills and abilities. The deep learning model could send similar matches to the employer, revealing job seekers that would make a good match for the open position.

“This model learns how people view jobs and better defines who matches specific job listings. We have open vacancies that have been difficult to fill for a variety of reasons. This deep learning tool helps us bridge that gap,” says De Blauwe.

VDAB has hosted PES leaders from multiple countries to share its success and allowed Radix.ai to demonstrate the project. Understanding the value of partnership, Radix.ai joined the AWS Public Sector Partner Program and is exploring the use of Amazon SageMaker as a tool to build, train, and deploy machine learning models at scale. Supported by the computing power of the AWS Cloud, the reach of the APN, and the success of JobNet, Radix.ai believes other European public employment services can benefit from this tool, just as VDAB has.

Learn more about Amazon Elastic Compute Cloud (EC2).

300x200_radix.ai_logo

Radix.ai, a Belgian machine learning startup, helps organizations innovate through the use of artificial intelligence (AI). Its goal is to help customers achieve strategic business goals by engineering AI and machine learning (ML) solutions that create competitive advantages.

APN-Standard-Consulting-Badge

For more information, contact Radix.ai on their website.