Community Comes First: Q&A with Streamlit Co-Founder and COO Amanda Kelly

Sometimes you meet early-stage founders who not only have technical chops, but also three other magic qualities: deep expertise launching products, repeat experience building companies, and a truly synergistic team mindset. That’s what GGV Capital saw in Adrien Treuille, Thiago Teixeira, and Amanda Kelly, co-founders of open-source startup Streamlit. After working together on AI and ML projects at Google X, the trio started Streamlit in 2018 with one goal: make it fast and easy for data scientists and machine learning engineers to create beautiful, performant apps. Made available for free download in late 2019, Streamlit has now has over 12,000 GitHub stars, has been downloaded over 1M times, and used in over 200,000 data science projects--from tracking COVID and analyzing NBA statistics, to predicting cryptocurrency pricing and providing automated cocktail recommendations. GGV led Streamlit’s $21-million Series A round in April of this year. 

I recently had the pleasure of interviewing Amanda at the Open Core Summit. We talked about Streamlit’s transformation from a tool Adrien built for himself to make it easier to code ML projects into one of the world’s fastest-growing open-source platforms. She provided lots of insights into how to foster fervent early evangelists, how to grow and support a massive global user community, and the best ways to design a commercial offering. Here are a few edited excerpts from our illuminating conversation.

Tell us how Streamlit came about. 

Our founder story goes back to 2012 when we were all working on a project at GoogleX. After that project got merged into Google, we all kept in touch. Adrien would call me up every six months and say “I have this great idea for a product and usually I would say, “That’s a terrible idea.” But at the end of 2017, Adrien was doing some modeling and didn’t have the toolsets he needed as a ML engineer, so he built the tools himself. I didn’t originally get it, but I knew if Adrien was using it himself, it had to be good. He showed it to Thiago, who immediately got it, and they both started coding the early version of Streamlit. In spring of 2018, they asked me to be a co-founder, since I have more of a focus on the business and product side. At first I said no, since I was happy where I was and plus I was seven months pregnant. But they kept trying, and when I was on maternity leave, they finally convinced me to join as co-founder and COO. I sat in on a call with Stitch Fix on how their data science team was using Streamlit; to see so much enthusiasm from such an admirable company that’s really pushing forward data science - that really convinced me to join.

Why did you decide to make Streamlit an open-source company? 

It was maybe an obvious decision but not an easy one. As technologists, we really all wanted it to be open source. But as co-founders, we knew we had a duty to our investors and employees to figure what the business model would be. Some open-source companies have done this really well and others haven’t, so we debated our business model for a long time. About two months before we officially launched the project, in October 2019, we rented a house near Muir Woods and said we would not leave until we had decided what exactly we would launch and when. We locked ourselves inside for 72 hours and then we took a really long walk in the woods, and by the end of that walk, we pretty much realized we had to go open source. That is what our users were telling us and generally what your users tell you is right, but we wanted to be sure. So when we came into the office on Monday and shared our findings, the engineers said “of course it has to open source.” And that was it, we decided right then and there on open source. Our employees and users know best. 

What downstream impacts has being open source had on building your team and instilling a company culture?

Our community just exploded overnight once we released this as an open source project. And that led to so many more opportunities we wouldn’t get otherwise. You get so much product and user feedback and that expands your ideas of where to go next, but you are also connecting with new developers every day. A huge part of our hiring pipeline is people who have reached out to us through the community. As a company, we have a hackathon every couple months where we all sit down and work together on the product. It’s really great that our employees are part of the huge Streamlit global community; we hire from the community and give back to the community. There is a lot of beautiful synergy there.

It’s clear there was a lot of excitement in the data science community when Streamlit went public on GitHub. What did you do right to ensure the initial big bang?

We had great early beta feedback so we knew we had something that resonated with data scientists. At the same time, you just don't know until you put it out there. We definitely planned more for it falling flat just in case, but we felt in our bones we had a real product market fit. No amount of marketing machine will counteract a bad product. One thing we did right in building our community is that we really tapped into an emotion people were feeling. All of us as founders had at least once been in a position where we felt kind of impotent because we were relying on another team member to build a tool we needed but it wouldn’t be ready for months. You just want to prove an insight and you can’t because you’re locked. Streamlit is really about unlocking the core creativity and intelligence of data scientists.

Another key thing we did was that we didn’t say there was one very specific use case for Streamlit. It’s easier to say “hey this is a bread for making sandwiches.” But instead we said “this bread can work for any meal.” That was a bit risky, but it fueled a lot of our growth because we were able to appeal to so many people. Within weeks we were seeing people from Major League Baseball and students in Nigeria and people at John Deere using the tool. People from all across the board were able to tell their own stories with Streamlit, and then they shared their experiences about using it online, which built the community.

During the last six months, you’ve launched functionality to let third parties create components that work within Streamlit and enabled sharing of apps among the community. What drove you to take this path?

Streamlit is so broad; we have people in every sector and in many countries across the world using the platform and that is amazing. But they all need slightly different workflows, libraries, or visualization engines. With such a huge longtail, we are never going to be able to create all the components needed by such diverse groups, so one of our first goals was to open our ecosystem to allow users to create components. One of the very first components we got was a 3D molecular viewer—not even in the top 100 things we thought would get created! But it immediately was a huge success with chemists and biologists. We welcome all kinds of components to be built for Streamlit to serve our wide community.

Streamlit has done an incredible job building a community. How do you track and measure it? 

Open source is difficult in that you don’t have the same level of data of a fully-wrapped product where you can just see where everyone has clicked. That’s by design, as we don’t want to violate our users’ privacy. Downloads give an idea of how you’re growing ,but they don’t show how you’re growing among key users, those who are embedding the platform into their workflows - the true evangelists. So we measure those users in a couple of different ways. One is looking at anonymous data to see how often people are using the platform. We also look at the data that’s missing, such as which operating systems we’re not seeing and why. On the community side, we use a tool called Orbit that gives us lots of information about the communities we engage with across Twitter, on our discussion forum, and elsewhere. We can see who needs help, and who might help us such as writing documentation or spreading the word about Streamlit. We also made a focused effort to create our own community forum. We didn’t go onto Stack Overflow because we wanted to drive and track the conversation ourselves. We also have a head of developer relations who is really focused on the data. We’re constantly looking at data to figure out how to make people in our community happier and elevate them to build the apps they want to build.

What are your commercial plans?

Streamlit is all about enabling developers to quickly create beautiful apps that help explore data. Once you build an app, usually you want to share it. You want to put it on a URL, share it with coworkers and the world, and that is difficult to set up securely. So that is what our commercial product will bridge. Creating any app with Streamlit will always be free open source, and we just launched Streamlit Sharing which is a free way to share apps. But later this year, we will start rolling out Streamlit for Teams, which will help developers bring Streamlit app deployment into their companies with enterprise-grade features around privacy, security, compliance, and collaboration.