What was the seed for embibe? How did you get interested in this problem?
There are two kinds of companies being built in the education world, and a lot of attention has been given to MOOCs. But if you look at why people spend on education, it’s largely because of employability. Employability leads to a better livelihood, a better quality of life. You don’t get to that kind of prosperity via MOOCS; you get there via cracking massively competitive exams (JEE is a good example). The world of exam prep has stayed largely offline, and it presents a very interesting opportunity. Technology coupled with data science can have a very meaningful impact on those exams.
What got me interested was the fact that I flunked one of these exams in 1999. It was basically because I did not focus on an essential element of the paper (chemistry), because I didn’t like it. But my liking or disliking a subject, or enjoying it, has no correlation with success in a cutoff-based exam. If somebody had corrected that blind spot for me, it could have changed my life. I was never a swotter, I always liked understanding what I studied, and for a person like that who’s engaged with study, to not get through because you’re blindsided, is unfortunate. This is one of the gaps that data sciences can solve, along with an intelligent recommendation engine.
How does the product work, at a high level?
It's supplementary to your coaching institute, or primary method of study. Time becomes scarcer as you’re progressing in your test prep, with school, tuitions etc all making demands on your time. Its important to know at any point where you stand. Even if you’re capable of getting a problem right, but you’re careless about the pace at which you attempt it, then that can be fatal for you in terms of the outcome.
We measure the speed of your attempts along with the accuracy , and we’re able to do it on a per-question basis, because of our use of metadata. We have an incredible amount of metadata about the question, and we store a lot of information about the student’s attempt, so the marriage of both these things give us the perfect insight about whether this was a wasted attempt, an overtimed attempt. We use expert polling to figure out the ideal time, and then we strengthen that with the data that we collect.
Teachers have an uncanny knack of knowing which kid will get into IIT and which will not. Our use of expert polling is an example of trying to extract the expert knowledge from their heads and encode it into our algorithms.
That’s interesting...so its not just data, you’re taking expert knowledge gathered over decades of experience and applying it to evaluate test prep work done by the student
Absolutely, because the ultimate computer is the brain, and our algorithms try to mimic that. We believe that our metadata is at the highest level of potency, in our judgement of where the kid is in his prep. The whole experience is gamified so you get sucked into practice and do more questions at a time. And we give you a lot of metrics around your practice. We tell you whether your time distribution and effort balance is optimal.
To be a good edtech company you have to be a good education company, and you have to really care about how the kid consumes information. We’re one of the few edtech startups that has full-time faculty on our rolls. The other thing we do is, we keep the community around the student engaged. Lets say there’s an effort imbalance - sometimes a student won’t do a subject because she doesn’t have natural interest, she’s bored, or she just finds what she’s successful at more interesting. Then you get the parent involved and say, hey, can you make sure that you get the student to focus. And the students respond very well to this kind of nudging - it’s almost as if we have a big brother/big sister kind of relationship with them, and they open up to us all the time about the problems they’re having. They send us Whatsapp messages, they send sweets to our office when they crack the exam - it is amazing to get that kind of feedback.
And finally, we use test attempts by people who go on to become top rankers, we extract heuristics out of their attempts on our tests, and use those to set benchmarks and hurdles for our good students. It's a continuously iterative feedback loop.
Can this become a generalized platform for learning beyond test prep?
The fact that you’re focusing on personal attention for a student using data sciences means you could basically create content around anything, because the data model allows you to scale based on level of difficulty. If you join the Khan Academy tree to this, you could create a learning platform that meets the need of primary to class 12. But i think we particularly are going to stay focused on the 8th grade on, as a startup we want to stay focused on that. One exciting use case beyond what we’re currently doing is, if the government decides it wants to solve the teacher shortage problem by using something like this. I think that’s exciting, and we’d be very happy to participate, have a conversation around that.
Can you envision a scenario where, let's say, I want an educational platform where I want really personalized learning, like a constant back and forth between me, the learner and the teacher, except the teacher is an algorithm, a really smart learning + feedback tool like Embibe that could almost replace a teacher?
At Embibe we really believe in the role of the educator. Which is why we’ve created an interface for educators, so we can get learning insights from students. We believe that education is an inherently personal thing, and part of that personal aspect is really the relationship with the teacher, and the persona of the teacher that drives you on. When you learn concepts, you want to bounce it off your friends and teachers. Classes are important not just because this is where you go to learn - classes are important because learning is human, and the interactions you have teach you to be a more well-rounded person.
I agree; I guess my question is more around trying to solve the scaling issue, and if Embibe can play a role, because I see an enormous unfulfilled need for teachers, and not enough good teachers to handle that scale…
Sure. We believe that educators have a really hard time dealing with increasingly large class sizes. The things that crowd a teacher’s time today - we want to take them off the teacher’s plate. Things like infrastructure, evaluation, assessments, all of them will be taken off. If they can tell at a glance, by looking at a student’s report that this is going to be a top performer. Set alerts to talk to students who might have a performance improvement issue, etc...that’s the future. There’ll always be teachers, they’ll just be doing different things. Smarter things.
Do you see this as potentially a global play?
I think we’ve got to get India right - that’s very important. If you look at the largest edtech companies in the world, they’re all global. The nature of the problem is global - if we make an engine that’s meant to improve performance, it will be useful anywhere. Once we get India right, and we see the right kind of traction happening here, we’ll be able to figure out where next. If we go a to a different market, we don’t have to resolve the content problem there - because we already have concepts, we just need to rejig them, provided the language of instruction is English. Education is a hard problem; anyplace there’s more people and less resources, its very hard. And the other thing of course, is that everybody needs it. That’s why we should go global.