Technology is becoming smarter. We want machines to be far superior than humans and serve us with minimal manual effort. Intelligence amplification or augmented intelligence (AI) is making this possible. AI referred as the effective use of information technology in augmenting human intelligence, is being drilled by technology companies to build smarter products.
Bay area based notetaking and archiving startup, Evernote is one of them.
Over the last couple of years, the five-year-old startup which has more than 100 million users, decided to have a separate AI division.
Talking with Siliconbeat last year during the startup’s developer conference, CEO Phil Libin had informed that the product which is in its “first phase” lets users search everything they’ve captured, is maturing to an “augmented intelligence” approach that will push stored information to you before you even know you have to search for it. “We want it to complete your thoughts,” Libin had said.
Today the two-year-old ten member AI team is being led by Zeesha Currimbhoy who had joined Evernote in 2012.
The Carnegie Mellon MS graduate in software engineering did spend more than three and a half years working with Oracle prior to Evernote. “Oracle was a great place to work but I always wanted to do more than that in terms of bigger impact. By my third year I realized that it was hard to make a personalized impact in such a very large organization so I decided to look around where I could use my skill sets,” Zeesha recollects over a conversation on Skype.
An early Evernote user since the time it was launched in 2008, the startup did make an impact on Zeesha. In fact Evernote was the only company for which she had interviewed for after moving out of Oracle.
Evernote Data Products team becomes AI
In May 2012, Zeesha joined Evernote as a data products engineer as one of the first employee in the AI team. Back then the team was known as Data Products and the work the team was doing then, as explained by Zeesha was “fixing the low hanging fruits.”
“My role initially was focused on designing and developing features for the product but as an engineer I was also thinking on taking Evernote to the next level, so that our users become more smarter. The role was the natural progression to my career with me leading the team.”
Evernote’s initial objective as we know has been to allow users to save web pages, snippets of text, e-mails, voice memos, photos and videos, but the key thing is also to understand the users and help them make smarter decisions.
“How do you build an app or an experience that really understands how you think, work and live? In the long run, we want the more you use Evernote, the better it understands you. We really want to be your external brain which could help you in making smarter decisions,” she explains on the rationality behind forming the AI division at Evernote.
The AI team
Initially, the team was just about a bunch of engineers who were setting up the base of the AI team to future proof the division going further. Two years back, Evernote had Mark Ayzenshtat as the VP of the AI team but starting this year he has moved out.
“We started with really small features that users expect like the typeahead search and search quality improvements,” adds Zeesha while discussing how the team has evolved. “As time moved on we brought in a bunch of really smart machine learning engineers, software engineers and formed an international team to support the kind of projects we perceived to do.”
With the AI team now forming its own vertical at Evernote, the startup hired a bunch of designers and QA professionals too.
AI projects at Evernote
The team which started with fixing smaller but much required features went ahead in building a Classification System. The project involved recognizing different types of content in a user account. Citing an example, Zeesha informs that one of the first application’s job was to identify without any input what notes or content in a user’s account are about recipes. The job of identifying content and making sense of it wasn’t easy but it also helped building up other apps in the same project.
Besides the AI team has also worked aggressively on Related Notes. “We wanted to help users when they were searching on the web. For example, if a user is searching for veg recipes over the web, Evernote would suggest the user that it already has a saved recipe of Paneer Tikka which she has not tried yet and might be a good idea for lunch,” shares Zeesha.
While this serendipitous discovery is interesting for the user, for Evernote, finding at the right time at the right context is the key. Evernote wouldn’t like to annoy users by showing content when it is not required.
Search has been crucial to Evernote and from keyword driven search the startup has been working to make it the way a user thinks or just type as they think, forgetting keywords. Today Evernote is able to understand the different parts of typed phrase and is able to pull out all the relevant information from the user account. “So it’s kind of query understanding based on how people think as opposed to what is expected of them to do. We want to remove the manual interaction between the app and the user and make it intelligent by bridging the gap,” she points out.
Descriptive Search, as Evernote calls it, has been implemented on the Mac client and for now it supports the English language. The startup has shared a descriptive post focusing on the behind the scenes walkthrough of the search methodology.
AI and Evernote Business
Evernote Business – the initiative by which the startup supports small and medium scale businesses has future plans to integrate AI in its offerings. Going further Evernote will support larger businesses which will bring forth the problem of information overload with increased users. The AI team is growing its capabilities from now to overcome the problems in future.
Talking more on it Zeesha tells me that suppose a user has a meeting at 5 PM on Monday, Evernote should know about it, the people who would be present in the meeting and it should also know the notes that would be relevant for the meeting. “Evernote should know this an hour before you go for a meeting. Making sense of this kind of data right across a company would be the very key for us to understand the people a user collaborates with or the groups of the people a user interacts with to make the offerings more intelligent.”
Further Evernote has plans for smarter offerings for businesses specially on mobile, which would mean making more sense of the data perceived.
Challenges and data problems
While Evernote is on a mission to provide smarter user experience by making sense of data, it is also thrown upon with a unique set of data problems. Zeesha shares the world might think that Evernote must be facing information overload and big data problems since it has grown to a hundred thousand users but the truth is that Evernote has a hundred million smaller data problems to deal with.
“At the AI Evernote team the challenge is how does one develop algorithms for a user who has just 10 notes or a user who has thousands of notes, in that case how do we scale our intelligent offerings for different sets of data. So at Evernote we don’t have one big data problem but a 100 million smaller data problems, which is quite challenging.”
This data problem becomes more interesting with Evernote’s support towards international languages such as Japanese, Dutch, German, Spanish, Arabic, among others. The AI team is dealing with such challenges on a regular basis and working its way forward. “The challenge becomes twofold when we need to make sense of data that is represented in a language which is not space separated,” adds Zeesha.
Going further the AI team wants to make Evernote as a smarter external brain. “Moving further I believe that we will build experiences that can be consumed on a smart watch and as the user enters a room, the television should carry forward the same experience building on your mental abilities. I would love to build products with connected intelligence that would take out a lot of manual work from a user’s life.” At Evernote things have started rolling so that the product could be aligned with wearable tech.
This may sound as a small beginning but as Libin warns software companies taking a wait-and-see approach – if you do so, you risk being left behind. There might be a chance that these intelligent offerings by the startup not only make the product smarter but help carve a new revenue stream for it. Just maybe!
But for now Evernote is making us smarter with the help of its growing AI team.