How difficult is it to build a machine learning platform? If you ask Google, you’ll find an assortment of bleak descriptions, including “laborious,” “slow” and even “nightmare.”
Senior Data Scientist Brenner Swenson can attest to these attributes. Among other tasks, he helps lead Reverb’s ongoing efforts to develop an internal machine learning platform; a task that requires not just developing suites of ML models at scale but also maintaining constant collaboration with other departments, including data engineering.
Swenson said that for this project, he has been working with a singular product team. Previously, data scientists were spread across various product teams, which made it difficult to collaborate effectively given the constant context switching. This shift has allowed him to develop stronger relationships with stakeholders.
“Having this new structural alignment has really fostered a sense of investment and accountability for the respective teams,” he said.
According to Senior Director of Analytics Cyril Nigg, this change is part of a larger evolution encompassing both the company’s data science and data analytics teams. Other aspects are shifting, too, such as the ways in which meetings are held and how team members collaborate cross-functionally.
In Nigg’s mind, having an effective approach to teamwork is essential, given the role data team members play in driving companywide growth.
“We’ve graduated from a scrappy startup to an emerging tech company where data inspires and informs the decisions we make,” he said.
We’ve graduated from a scrappy startup to an emerging tech company where data inspires and informs the decisions we make.”
Considering how important data is to the company’s core operations, it’s crucial to ensure that the quality, documentation and organization of this data improves over time. To do this, Senior Marketing Data Analyst Melissa Mascarenhas helps maintain the data visualization tools used across the company, including Looker and dbt Labs.
Since joining the company about a year ago, Mascarenhas has not only been tasked with maintaining the effectiveness of the company’s data initiatives, she has ensured her team’s culture is defined by transparency. By instilling helpful practices, such as regular stand-up meetings, she has enabled team members to connect openly during the workday.
“It’s really important that we make sure we check in with each other as human beings to feel like one team,” Mascarenhas said.
This idea of building ‘one team’ is critical to empowering what Nigg calls “data-inspired decision-making.” Through a combination of collaboration and a team culture that fosters honesty and fun, data team members at Reverb get to play a direct role in shaping major decisions while gaining valuable experience along the way.
“There’s a lot of interesting data and problems to work on, and we have a direct impact on the business,” Nigg said.
A GLOBAL COMMUNITY OF MUSICIANS
A wholly owned subsidiary of Etsy, Reverb offers an online marketplace dedicated to buying and selling new, used and vintage musical instruments. Buyers and sellers across the globe use the company’s platform to exchange all types of musical equipment, from drums and guitars to keyboards and synths.
Collaboration Fuels Innovation
If data is key to impacting Reverb as a whole, then the machine learning platform its team is building will form the heart of this influence.
According to Swenson, the development of the platform has enabled the data science and data engineering teams to work with the latest tech stack components and migrate off of the company’s previous platform; a process that’s far from simple.
For the past several months, Swenson has been migrating machine learning models and replicating them without degradations, while also adding some improvements. This feat, coupled with writing lots of code, enabled his team to create an improved version of their price recommendation models, which help sellers gauge the right listing prices for their items.
This undertaking is essential to driving progress at Reverb, which is why it’s especially important to ensure the data being used is reliable. For that, Swenson and his peers rely on Mascarenhas’ team, who work to continuously improve the quality of the company’s data.
Mascarenhas explained that her team implemented checks to ensure that whenever someone adds a new metrics or data source to Looker or the data warehouse it passes certain requirements to maintain a higher level of quality.
“We spent quite a lot of time figuring out what those parameters and dimensions should be,” she said. “It also took a lot of time to clean up the environment to ensure that everything was up to compliance.”
Once compliance was met, Mascarenhas’ team ran tests and submitted dummy requests, ultimately ensuring everything was working properly. Not only has this work enabled her team to continuously educate employees on the importance of parameters, it has also allowed them to review documentation.
“This collaboration enabled us to provide important documentation to which people can continuously refer and follow tests that adhere to our standards and the quality that we’re looking to maintain,” Mascarenhas said.
The documentation created by Mascarenhas’ team has been a game-changer. Besides offering team members a deeper look inside the company’s data operations, the team’s work has saved time for individuals across the organization.
TECH CHOPS AND BEYOND
Working on one of Reverb’s data teams requires more than technical chops. Swenson shared that strong data science candidates possess curious dispositions and a scientific mindset. “You have to have the quantitative know-how to take your interactions with stakeholders and turn them into actionable insights or analyses,” he said. Swenson added that the company’s data scientists are also expected to write high-quality Python code.
For Nigg, it’s equally important for data team members to bring diverse perspectives to the table. “I’ve learned a lot from everyone on the team. We have a great mix of people from a wide range of backgrounds that can get you thinking about a problem in a way that you hadn’t thought of before,” he said.
Space for Empathy and Connection
Although working on one of Reverb’s analytics and data science teams involves hard work and endless collaboration, that doesn’t mean time for interpersonal connection and understanding is pushed aside.
Mascarenhas noted that the company has made an effort to bring mental health conversations into the workplace. In a recent meeting, she witnessed some of her peers discussing the “well of emotions,” which helps indicate an individual’s current state of mind.
“I shared this with a friend of mine who’s a counselor, and she was like, ‘Oh, you talk about that at work? That’s breaking boundaries,’” Mascarenhas said. “The importance of our team’s mental state is something we need to take into account as coworkers, but especially as leaders.”
The importance of our team’s mental state is something we need to take into account as coworkers, but especially as leaders.”
Mascarenhas isn’t the only one who feels this sense of openness. Nigg said he has never worked at another organization where team members were encouraged to have deep conversations during the workday.
Naturally, given the supportive nature of Reverb’s work environment, team members take time to connect on a personal level. Swenson and his peers often play the geography game GeoGuessr and take part in hackathons.
During the most recent hackathon, data scientists were paired up with others from across the company with whom they don’t typically interact. Swenson had the chance to put together a project alongside members of the data engineering team, which proved to be very impactful.
“It sparked a lot of great and interesting ideas,” he said.
Whether they’re busy tackling a project or laughing together over a game of GeoGuessr, data team members are always leaving their own mark on the company.
For Nigg, this influence is bigger than many people may think. Not only do data team members get to drive the company’s mission “to make the world more musical,” but they get to do so on an international scale.
“You get to see users and interactions from people all over the globe,” he said.
Reverb’s data teams are the heartbeat of the company; a truth so palpable that it compelled Mascarenhas to join the company in the first place.
“If you’re looking to make an impact, Reverb is a great place for you,” she said.