Who’s this other Rishabh?
Some weeks back I did a podcast with Rishabh Mehrotra - an AI leader in Europe who is currently the Head of AI at Sourcegraph, prior which he was leading AI at ShareChat and was a staff research scientist at Spotify. He’s got a PhD from UCL in Machine Learning and a double major from BITS Pilani in CS and Maths.
I’ve known Rishabh for a year now - and he’s one of the foremost leaders in applying ML for marketplaces, and is passionate about building teams.
The big takeaways from the podcast are below:
Europe and UK have a thriving and growing AI landscape
A lot of investments are going to the infrastructure and people are now switching to applications. Poolside, Stability, Mistral from the model world, Synthesia for video generation, Photoroom for image generation. There’s companies like Graphcaore that’s been building IPU (though sadly had an unfortunate acquisition recently), Wavye (autonomous driving) and many more.
Europe has had a rich history of AI (right from the days of Alan Turing) which Spotify, Apple, Meta and Google all having significant AI teams in the continent. However, Deepmind has been the backbone of this movement - whose alumni has gone and started a range of companies.
Almost a decade ago Amazon Alexa acquired a Cambridge startup called EVI technologies, that were developing graph knowledge bases, which were used for intelligent conversations. Spotify also acquired Nylon Labs from Paris.
Advantages of European AI talent
Rishabh says that because Europe has had a rich history of building global products at scale - be it via homegrown companies like Spotify or via big tech.
Eric is one of the founders of Model Labs and he was the original person to develop Annoy (Approximate Nearest Neighbours). There's a long history across each of these companies. So you start seeing a lot of amazing talent who have done it at scale, which is the important part. So I might invest in them versus somebody else because they have seen these problems at 100-200M MAU. And that experience at that scale is something which counts a lot.
And so it’s a great place to be hiring senior talent that has seen the scale of say 100M DAU. The big tech talent in India tends to be specialized on specific areas. For example the engineering teams at Google, Walmart, Amazon, might not be working on their home page
(e.g WalmartLabs in India might not have much experience with their consumer facing website, but might be excellent on their merchant technology initiativesIndian domestic ecosystem has
Why and How ShareChat Built a Global AI Team with a Focus on Europe
Rishabh credits ShareChat’s founder Ankush for his conviction to build a global AI team. The reason, he says, is because Ankush realized that ShareChat’s new product Moj was a feed product - and they just had to crack the recommendation engine for the product to do well. Finding senior talent that has that experience was hard in India and they were convinced that they had to find the best talent regardless of where it came from. Rishabh along with Debdoot (ShareChat’s Head of AI) surgically build a stellar team of senior engineers across Europe and US, which complemented the team in India.
Yeah, I can talk about a bit composition, right? By the time I left, I had a team of 60 plus. it was a mix of 50 to 55 % like SDs, almost 30 % MLEs, and then the remaining like scientists. I really prefer having at least like a portfolio where then like, you can have like at least 5 to 10 % of your team is like applied scientists who can really innovate on the edge because you need them In terms of the leveling and where they were, I mean, some of the senior staff engineers were not just in UK. I mean, like we had some people from TikTok in Poland. Why? Because they've done some of these computer vision problems and like rendering and a bunch of those at scale for TikTok. So let's hire them. We got quite a few people in different parts of Europe and also in the UK. We also had few people in Seattle and one in New York. I think this is where we were very intentional about, look, this is a topic. Where do we bring the right talent for this topic?
A hiring experiment that worked
(At ShareChat) in terms of the backend systems, in terms of the machine learning systems, the stuff they are building right now is really impressive.
The ARPU in India is low. What that means is the cost of recommendation per user has to be low. Now, this is not something which you get for free, right? If you're serving X million users, daily active users for your recommendation system, I think ShareChat is probably in the top one or 2 % of companies having the lowest cost per recommendation per user essentially.
And this is not just machine learning achievement. This is like a huge system scale achievement. Because, I mean, we are paying like higher, I mean, double digit, higher millions to GCP for our cloud. And we reduce it by more than 50 % in the next year and a half. And you can do this if you have the amazing talent.
India is great for junior AI talent. Senior talent is a challenge
When I was at Spotify, one thing which I was missing was very high talent density of junior talent, which I can hire quickly and like they can really learn on the fly. At ShareChat, we had somebody who joined just two years after undergrad in India. And we deployed a contextual bandit model, contextual bandit for people who don't know it's like step one reinforcement learning. And that's what we're using for ad load balancing. Like how many ads do you show on the feed? If you show more ads, people are going to churn.
Now, an undergrad with two years' experience who did not know Bandits or counterfactuals on day one, within a quarter and a half, he was able to deploy it across 60 -plus million daily active user scale. And this is something which I hadn't seen before.
So I'm very bullish on like the junior talent in India
But at the same time it was a nightmare to hire like senior staff engineers or staff plus senior staff and principal engineers in India. They would be at like the Google Maps or some at Microsoft. But not a lot. And that too in very specific topics. Not all the topics.
US has great talent in ML and across all topics, but it comes at a cost.
India has great talent in pockets - but it’s hard to know about them
Dream11, and many other companies have great talent but you won't know them. And I really tried poaching a lot of people from some of those companies. They have like amazing experience, like artifactual evaluation, LTV calculations, but then they don't talk, they don't publish, so hard to know about them.
Hidden gems in Europe
Europe has great talent across, I mean, the Ukraine war with Russia was unfortunate, but a lot of amazing talent from Yandex (with great production experience) was fleeing Russia. And a lot of people were in Turkey and they just landed in UK and they were looking for jobs. And it was at one point, it was cheaper for us to hire in Europe compared to India.
Europe a great place for Indians
When I did my PhD, I started a company. We got a deal of like 75 ,000 pounds, which is amazing at that point of time. But I didn’t have something like the Global Talent Visa. And hence took up a job.
So there was an opportunity for me to start up way back in 2017. Today I'm on a global talent visa. And I know you've helped a lot of people that obtain global talent visa. I keep pushing this to a lot of people. A bunch of really smart Indians are now on global talent and are free to startup
I mean, one of the reasons why I didn't move to the US was because of the immigration system. That is not the case with Europe.
Advice to founders in India on building engineering teams in Europe
I think the first acknowledgement is that you're not going to hire the masses in Europe, right? Again, you're selectively going to hire only key strategic roles. So if you have to get a team of 30 people, you're going to hire like 28 of them in India still. So your biggest hiring headache is India, sure. But if you're a good leader, then you would understand the pivotal role these two or three talent plays. And also this compounds over a period because the junior talent that works with them, will become senior, staff engineers and so on.
So don't just look at it that, yeah, I'm going to invest this much and get these two people today. And what's the impact for my metric in the first quarter?
Advice to founders in US on building engineering teams in Europe
You always have a budget, right? I spent a lot of time with my CFOs. I’d tell them “Hey, why should we put notification, ML on notifications? If we spend X thousand dollars, you're going to get this ROI and we're going to land you the money in like X months from now. I was like very capitalistic in my thinking that look, you spend this much money now. I can make this much money six months from now.”
And here's the ROI deal, right? Let's handshake on that. So that learning was great. I think like to me, it's more about like, hey, if you have a budget and if you want to optimize the best out of that budget, then it makes sense to kind of consider like Europe for sure.
Building global AI teams w/ Rishabh Mehrotra - Sourcegraph, Spotify, Sharechat