An App to Classify the World’s Art
From day one, the team at Art.sy has never been short on ambition. The team seeks to expose as many people as possible to art by building “the Pandora of the art world”. They’re enrolling the help of high profile investors and advisors such as Peter Thiel, Joe Kennedy, the CEO of Pandora, and powerful New York city art dealer Larry Gagosian.
Art.sy’s “Art Genome Project” currently classifies 25,000+ artworks by over 4,000 artists. Just a simple search for “Van Gogh” on the site returns similar artists based on 800 different characteristics, called “genes”. Despite having steep technical requirements though, no one at Art.sy actually manages servers. By leveraging Heroku and powerful Add-ons such as NoSQL database MongoHQ, Art.sy was able to focus on building a solid recommendation engine rather than on servers and infrastructure.
100 Percent Focus On User Experience
“Working with Heroku has changed the requirements,” says Art.sy head of engineering Daniel Doubrovkine. “I don’t need to hire operations people – the person who carries a pager for when the server hangs. I don’t have anyone carrying a pager. This is an entire new generation of operations that changes how you structure your teams.”
Focus is at the core of Doubrovkine’s decision to use Heroku. “To me, it was logical to use Heroku. If it’s my core competency, I’m going to build it. If not, I’m going to buy it somewhere,” says Doubrovkine. “I’m not inspired by building the most horizontally scaling architecture. Setting up caches, proxies, front-end routing, all of that is stuff I don’t want to deal with. Once you’ve felt the pain of managing your own servers, you don’t need to feel it again.”
With operations being managed by Heroku, Art.sy set its sights on getting the top interaction designers, front-end developers, and mathematicians. They also employ some of the foremost experts in the art field, including many former employees of public institutions, auction houses, galleries and museums.
Deploying New Features at Scale with Confidence
When Doubrovkine’s team expressed concern about his plans to launch a new set of features to an influx of users anticipated from an upcoming feature in the New York Times, he explained everything would be fine.
“The Heroku approach is different from the old school where you had to measure very very carefully the kind of traffic you could take. You used to hope that your math was correct and usually it was not. With Heroku, I actually don’t have to rely too much on guesswork. I’m not constrained by hardware, the capacity of the network, and all those things that are hard to get in physical or even virtual hardware.” -Daniel Doubrovkine, Head of Engineering, Art.sy
When the story finally broke, the team flipped a switch on a whole set of features consisting of more than 15,000 lines of code. Art.sy’s traffic immediately multiplied by a factor of 15. As time went on, the number of simultaneous users on the site grew, but Doubrovkine explains that they simply increased the number of dynos to keep everything running smoothly.
“We can handle traffic spikes with the infrastructure we have, Doubrovkine says he explained to his team. “If we need 50 machines, we can have 50. If we need 2, we have 2.”
A More Efficient Workflow
Doubrovkine points to Heroku’s extensive list of Add-ons and the ability to instantly create multiple testing and staging environments as yet another reason to move on from the traditional way of doing things – where you need to contact an operations team every time you want to test something out – and on to Heroku.
“The ability to create more environments in that same infrastructure is absolutely fantastic. Heroku treats every app pretty much the same, so there’s no segregation of features. The fact that all Add-ons are available across the board, is very important,” says Doubrovkine. “The fact that you can get all these pieces provisioned and connected to your app is very powerful.”
He says that he often shows off Heroku to other CTOs – people responsible for managing large technology organizations in larger corporations – and that they are amazed at what they see.
“When I show them how we operate on a daily basis – the process of taking code to production, all those magical things that just work – their eyes open wide and they think that I’ve come from another planet,” says Doubrovkine. “Our process of taking code to production and the Add-ons that ‘just work’ is pretty new to them, while I almost take Heroku for granted.
This is very much the way we should develop and run web services. It is an entire new generation of operational workflows.”