Walking into an office building and handing the receptionist your resume is no longer enough. Today, millions of job seekers rely on digital platforms like Indeed, JobSeeker, and FlexJobs to find jobs that fit their skills and rapidly apply for multiple opportunities.
However, if these platforms don’t show relevant opportunities to job seekers, both employers and candidates miss out on ideal matches that could provide optimal ROI for organizations and fulfilling work for employees.
Ensuring this accuracy and relevancy is the job of software engineers like Manjunatha Jagalur, who leveraged his deep experience in machine learning to refine the search engine for one of the largest job hunting sites globally and serve the right opportunities to the right employees — without sacrificing a great user experience in the process.
Building ML models to find the right job match
Job seekers might look at job searching platforms and see them as nothing more than a keyword-driven search engine, but matching candidates with relevant opportunities is much more complicated than that: For these types of platforms, the engine needs to account for hundreds of unique factors like location, skills, tenure, education, licenses, certifications, and other attributes.
Software engineer Manjunatha saw this problem as a balancing act: “The end goal of most job search engines is twofold: optimizing for end user benefit while providing good ROI for the advertisers,” he says. “The main problem is balancing these goals.”
To this end, during his time as a software engineer at a large job-hunting site, Jagalur designed what he refers to as an innovative composite ranking architecture. This was essentially an infrastructure that combined multiple machine learning models into a modular pipeline that could be tuned for different goals as strategy evolves. By introducing these tunable parameters, the platform could adjust optimization weights to enable better user matches while still improving ad targeting outcomes for paying clients.
Optimizing for placement without sacrificing speed or fairness
While capable of improving job matches and successful hires for employers and job seekers, the model still needed to be optimized to keep the main, user-facing search engine snappy and user-friendly.
“Increased latency drastically reduces user engagement,” Jagalur says. “So it’s crucial to have a low-latency system because even small increases in search response time could send users to other platforms.”
This focus on optimization required Jagalur and his team to break down the machine learning process into multiple steps, pre-processing data wherever possible. That meant only a small selection of matching models would need to be applied, like specific job titles entered, at the actual moment of search. Other parameters, like existing skills and experience, could be analyzed and assigned to user accounts before searches were even made, shortening the processing time required to provide high-quality matches.
To ensure a fair and unbiased search experience, Jagalur’s team created tools that transparently showed how tuning different model weights would impact results. By making these outcomes quantifiable, business strategists could tune models to produce more diverse applicants or show job-seekers a wider range of opportunities if other parts of their profile indicated a potential fit.
An ML-powered marketplace connecting millions worldwide
The result of Jagalur’s work is a tunable, scalable system that offers precision, fairness, speed, and strategic flexibility to a marketplace that now connects millions of job seekers and employers around the world.
This wasn’t just a model refresh; it was a full-scale redesign of the platform’s decision-making engine. Today, millions of people are finding meaningful work through the platform, thanks in part to the robust job matching system and highly responsive user experience that Jagalur helped create.
VentureBeat newsroom and editorial staff were not involved in the creation of this content.
