How a laid-off quant put big data to work for job seekers: Roletroll

The idea behind Roletroll is that computers can find job listings better than you can. The Boerum Hill startup uses machine learning to match language in résumés with language in job listings.

The site, built by Adam Grealish and Ryan Buterbaugh, has been live in New York, Chicago and San Francisco for a little over a week. 

“There is enough information available now that job seekers no longer need to manually sift through job listings,” Grealish said via email. “Applying big data techniques allows this process to be automated, bringing the best job matches to the forefront. This gives passive job seekers more time to excel in their current role and active job seekers more time to be active on other aspects of their search — networking, courses, meetups.”

Grealish, a former quant, built the basic idea behind Roletroll after getting laid off from a big financial firm in 2010. He used the application to land a job at Goldman Sachs, which is a strong proof-of-concept. Later, he decided to really focus on building it out, according to Business Insider.

It’s still very early in the company’s development. The two cofounders have bootstrapped Roletroll so far, which is currently only available to software developers and finance professionals.

Roletroll’s founders say the service will likely stay in only those two sectors for the time being. “We plan on iterating to a well defined and replicable product in the finance and tech space before branching to other industries,” Grealish wrote. “So you will see more locations come online within tech and finance first. But outside of fin and tech, healthcare is probably the next industry we would focus on.”  

The site primarily uses open-source tools, largely grounded in R. Roletroll uses big data analytics to turn the language of job listings and résumés into data. After that, the company’s software can analyze the data and turn it into good job matches.

Roletroll features two broad algorithm sets that drive user results. Contextual match algorithms draw from Apache OpenNLP, a program for analyzing natural language text. These sets make sense of résumés and listings and decide whether any given two are solidly related to each other.

The second algorithm sets are adaptive. These use feedback from the user to refine user results: think thumbs up/thumbs down. “We use a battery of proprietary and machine learning methods to achieve this,” Grealish wrote.