Back
Oct
25
2017

Interview with Vasco Pedro, co-founder of Unbabel

  • Image provided by:

Vasco Pedro is a serial entrepreneur who is currently devoting his attention to his language-tech startup Unbabel. Vasco will be speaking at this year's Digital Freedom Festival in Riga.

Originally from Portugal, Vasco's ambitious team has participated in the USA's YCombinator accelerator, and recently launched a language-accessibility translation program under the TechFugees initiative.

We sat down to talk to Vasco about the early stages of running a startup, and gaining global recognition while operating out of a small European country - Portugal.

Where did the idea for Unbabel come from?

It all started during a surf trip with my soon-to-be co-founders. We had been surfing together for about a year before founding Unbabel, and we would always talk about doing something together. We all have PhD's in computer sciences and specifically natural language processing, and so language was always a passion.

We kept thinking about this really compelling problem that we were seeing, which was how hard it was for people to seamlessly communicate in different languages. Especially in situations that we were seeing at the time, like people having airBnB's and people from all around the world coming to their houses and it was really hard to interact with them.

We thought; “It shouldn't be this hard anymore, it's 2013!”

What's the solution?

If you look inside a machine translation engine, at the time, it was praise-based (now most engines are neuro-based), there was a graph that the engine used to determine the right translation. It would pick from different alternatives, based on probability. But those probabilities a lot of the time were mis-adjusted. So we thought - what if we can add a human element to that? What if we had a symbiosis between human and a machine? Where humans are providing a qualitative output and teaching the machine to be better, while at the same time providing professional grade translations.

We took machine translation engines and combined them with humans - specifically non-professional translators and then built an AI to enable them to be super-translators. Bilingual, able to produce high quality work in a super-scalable and fast manner.

What's the endgame?

I think in the foreseeable future, humans will be involved in some way. I think in some use cases you'll start seeing MT (machine translation) being more useful, but we're still far from that being enough when it comes to human communication.

The role of human might change over time, it might go from post-editing, to data training, data generating to train the engine.

And then texts is just what we're doing right now. There's still speech, one fundamental problem that we're trying to solve is that people speak different languages, and that's not going to change. That creates a lot of problems for businesses. Any business that wants to work in multiple markets has a language barrier. Solving that is going to be very important.

The ultimate endgame is where Unbabel is embedded in every interface, tool, screen in a seamless way that's pretty much invisible.

I saw a metaphor recently, in a TV show called The Good Place. The scene is that there are two characters in heaven, and one of them says “Wow, you speak such good English!” and the person replies “No, I'm actually speaking French, but there's something up here that is continuously translating everything back and forth so that it seems like you're speaking French to you and I'm speaking English to you.”

That's kind of our vision. A moment in time where people can easily communicate - it's any-lingual, it's language-agnostic. No matter what language you speak you'll still be able to travel and do business with any other human.

You're from Portugal, but experience in the US. How did that work out?

I lived in the US for about 10 years. I went in 2001 to study my Masters'. Two years turned into 10 years, 3 kids and a PhD. Most of my time I spent in Pittsburg but traveled around, mostly doing internships.

I came back to Portugal in 2011 and by then I had started a startup in the US, and it merged with another startup and the idea was to start a team here. Then I co-founded another startup, that didn't quite end up well, and then there was Unbabel.

You went to SF, but came back, why?

There were several reasons.

One was personal reasons - I was supposed to be out for 2 years but ended up staying for 10 years. After 5 years I was feeling like an immigrant and I really missed Portugal and I had family there. By that time I had kids - I wanted my kids to experience this part of their culture, and have a relationship with their grandparents.

Second, at the time Portugal was going through austerity, and I believed it was an opportunity to do really cool things. It was a time where I would have more impact. This was right at the beginning of the startup movement in Lisbon, and people were already thinking, you know, I don't have a job, so I might as well try something new. People were less risk-averse, which made it easier for a lot of things.

A lot of the great companies start in recession periods, and there's a couple of reasons for it;

  • Companies have access to more human resources - less people have jobs, there's a bigger pool to choose from and you can pick the cream of the crop
  • Because you have to start and survive lean, so in better times when the recession is over, you're well positioned to take advantage of that and leverage what you've already built, while everyone else is just starting
  • People are less risk-averse - they're more willing join a freshly established company and team.

What was your experience in YCombinator, and what would be your advice to other European startups?

We founded Unbabel in August 2013, and in October, we decided to apply for YCombinator. A little bit out of just “why not”? We knew about YCombinator, thought it was a good idea, but no startup from Portugal had ever been admitted.

Just doing the application forced you to answer some questions and look at your business from a critical perspective and make some fundamental decisions.

In November we got a call for the interview, the process is intense. It's a 10 minute interview, all founders have to be there for the interview, and you're given rapid-fire questions. And then by the end of the day they make a decision - you're either in or you're out. It's very efficient.

In January we moved to San Francisco, did YC, raised around, and then we decided to come back and do the headquarters here.

I certainly recommend the experience. Even the application process is useful for startups.

Would Unbable have been successful without YCombinator?

Definitely not. Getting into YC tripled our valuation.

The day we were about to sign the term sheet for our first investment round was the day we go the email from YC that they wanted to see us for the interview.

Between getting into YC and leaving, our valuation multiplied 7 times. So basically we would have had to do the same thing, but with much less capital, which would've been hard.

The actual value is much more than that. There's a very no bullshit environment in YC. And they only way to get shit done is to do it. It's a great excuse to be very productive for 3 months, and do what it would've otherwise taken you a year to do.

Either you crumble, or you become more valuable.

Any advice to startups to make the most of accelerators?

I want to mention a caveat - I don't feel like I've done enough to be qualified to give advice, I know more about the things that didn't work well with me, but every story's a different story. But if I had to name advice, this would be it.

The first bit of advice is commonplace but still very true - the most important is the team. But a team that pays off their emotional debt.

A big killer of startups is emotional debt between founders.

You need to prioritize spending time with your co-founders outside of work. Those bonds are what will carry you through the hard times. It's easier to have co-founders, because there will inevitably be times when you are down, and hopefully when you're down, then they'll be high. That way you'll support each other.

But there's no substitute for hard work in the beginning. At first you need to execute more than you plan. Only when you have data, then you can start to become more strategic in your decision making.

Second, it's more of a common sense thing - you can build your tech forever until it's perfect, but customers are more important. Until you get your product in front of customers, you won't know if it really works.

And third, and this advice used to irritate me, make sure you have great investors. If you have the option. It does make a difference. A shitty investor can destroy a company.

To identify a good investor, look at their track record - talk to other companies that invested in them and ask them what it was like. I like investors were founders themselves. Having done that helps you understand what goes on in the minds of the founders and can give you some shortcuts if they've done them before.

There has to be chemistry. You have to like them. If not, neither party gets much value.

You were involved with TechFugees, can you tell me about that initiative?

From the beginning we always thought that, even though we're tackling B2B SaaS, the goal is to help the fundamental problem. We always felt that there are others who are more in need of access to translations, like NGOs. We didn't have the resources at the time to do it. But once we had a big enough translator community, we sent out a questionnaire to the community asking them if they're be willing to volunteer part of their time to help NGOs. The response was overwhelmingly a very big yes. We were amazed at how many people wanted to participate.

At the time we talked to Mike Butcher who is very involved with TechFugees, and though that this is the time to get involved, and launched Unbabel.org as a part of it.

We started to work with NGOs. An NGO signs up, and people can sign up to work with that NGO. Whatever translation work comes, we provide the infrastructure and tools for free, and our translators provide the translations. This way we can offer translations to NGOs for free, while still maintaining a high quality.

It's still very much in the beginning and we're iterating to see how we can make our community more efficient. But the will is there, on both sides.

We're very proud of that.

What are you currently focused on?

We're looking at how we can enable companies to offer multi-lingual customer support in a simple way. A lot of times companies don't provide customer support except for in English. That provides a bad experience for the rest of the world.

If you look at the statistics, a lot of the new users coming into the internet are coming from the developing world - the biggest language is not going to be English.

We have native apps for a lot of the customer support services like ZenDesk, Salesforce and other helpdesks. That enable companies to offer support in 28 languages. It works as a layer.

A support email, with all of the human element, takes us 5-10 minutes. For email, that's acceptable. For chat, we expedite the translation process, 80% of stuff is machine translated, and what isn't, takes about 1-2 minutes to be translated. It doesn't affect the flow of the conversation.

You're coming to Riga for DFF this November - what will you be talking about?

It will be my first time in Riga - I'm very curious to see the city. I'll be talking about AI and the future of where humanity is headed.

I'm very much in the field of AI - I believe humans and AI will merge, and that we're going to become the AI. I think the cross has already happened in the sense that very few people can work nowadays without a computer, there's a limit of our brain being dumped already into computers - how many phone numbers can you remember? I think that's going to continue to happen through different projects like Elon Musk's Neuralink, I think that 10 years from now, this will be the future.