Xavier Anguera
ELSA co-founder and CTO

When I joined ELSA Speak in 2015, I was tasked with the challenge to build and lead the engineering and research teams at ELSA. My professional path until that point had been focused on research in speech and multimedia technologies, both in academia and corporate. This new challenge has been so far both a great learning experience and a hell of a ride!

In this post I highlight the main differences that exist, in my opinion, between the research team we have built at ELSA and some of the realities I have been able to observe so far during my professional career. I am aware that both academic and corporate research have many implementation variants, all of them valid if the objectives they set themselves are achieved.

At ELSA we were set from the start with the objective to build the best English language speech assistant for people that feel shy to practice English with another human or do not have the opportunity or means to do so. Our goal was to build an accurate system to detect what pronunciation/intonation mistakes users make when speaking English, and to help them improve by offering relevant feedback. Building such a system was not a trivial task, and I don’t think we are done yet, but what I found most challenging (although also very interesting) has been to always keep a balance between pure research and continuous delivery of results.

Building a product out of your research at ELSA has been quite different from the research I was used to in universities and corporations I have been lucky to be part of. Research in a small startup means, among other things, that you need to choose very carefully what problems you want to solve, making sure that you can make good progress in a reasonable time. In addition, you need to continuously measure your progress and be able to quickly switch gears if you see the project is not going in the direction you were initially expecting and is stagnating. Money in a small startup is a finite resource, and your research is likely to have an impact on the near/mid-term future of the company.

In addition, here are some other differences I have observed over the years:

• In academic research the main goal of the researchers is to advance the state of the art on a given topic and, thanks to it, publish their work in top tier conferences or journals. This awards the researchers and their team some prominence in their field and will make it easier for them to get new funding to continue with the research. In a startup like ELSA, we are also interested in pushing the state-of-the-art forward, but rather than getting kudos from peer researchers, we are driven by the many users that get to use our new algorithms or improvements we make to the core. More so, instead of measuring our success using impact factors, we measure it with startup KPI metrics such as engagement, retention, virality or conversion.
• In some big corporations the research labs are created to “impress” the company stakeholders (or the public in general) by providing a showcase of how innovative the company is or can be. Unfortunately, many of these innovations end up locked in a drawer and never get to see a real user. On the contrary, at ELSA our researchers have the power (via a single click on our CI/CD pipeline) to deploy their code directly to production, impacting millions of users of the ELSA app right away. This great power comes with great responsibility and this is why we put in place good internal processes around automated code testing and manual code review among researchers to make sure that few(er) bugs make it through. Each and every one of our researchers at ELSA feels very close to the end user and is very aware that their research has a direct implication on many people that want to improve their English.
• In some big corporations that do turn research into products the process is sometimes quite slow, spending multiple quarters to go from pure research to tech transfer and finally productization. In these cases, the end product reaches the final user many months after the researcher finished their algorithm. The advantage of a small organization like ELSA is that projects’ lifecycle is usually quite streamlined from concept to deployment. The inception of a new feature either comes from the research team’s realization of some possible use case or improvement for the algorithms, or from the product team with a request for a feature that they believe we should offer to the users but needs some new technology. This allows our researchers to stay razor-focused on performing research on the topics that can impact the ELSA product the most. I should note that there are some big organizations, like Google or Facebook, that have been able to retain the startup mentality in how they go from research to “reality” (i.e. a product) and sped up the research process to make it as similar to a small team as ours as possible.
  

Our researchers at ELSA actively participate in the whole lifecycle of the product, starting from the conception phase, discussing with the product team about the features to be built, all the way to the productization, interacting with developers on how to integrate the research into the apps, and finally testing and monitoring of its performance. This generates a strong feeling of ownership of the features we make possible through our research and a sense of accomplishment when we see people use them and we help them improve their lives through language education.

In 2020 we are planning to expand our speech research team. We are looking for people with a PhD in speech/AI and some experience building products. If you like our ELSA app and you believe you would be a good contributor of our research team, send us your CV at [email protected]