In this podcast, we spoke with Alex Weidauer, Co-founder and CEO of Rasa. Rasa is a Python-based open-source conversational AI framework powered by Rasa NLU and Rasa Core. We learned the backstory of Rasa, which was thought up in Alex’s kitchen while building a chatbot using Dialogflow and other chatbot frameworks — till today with 3M+ downloads, and 10K+ members on the global Rasa community forum.
We also touched upon the story behind naming Rasa, the latest GPT-3 NLP research, talent acquisition, product-market fit for startup products, recent Series B funding from Andreessen Horowitz(a16z), Rasa Masterclass on YouTube, and even Alex’s favorite Bollywood movie and actor!
Theme Introduction: In this podcast series, we will have a light and casual chat with Founders, Tech Leads, Hiring Managers/HRs, Data Scientist, Researchers, Community Builders about how they use AI to solve real business problems. This podcast will bring you closer to all these amazing companies and get you excited to work with them. So don’t forget to stay tuned to all the upcoming episodes of the series and make sure to subscribe and follow us.
Alex likes to float on the beach 🏖️, eat salads, and loves coffee ☕ which goes along with communities 👥
Big thanks to Yogesh Kothiya for concisely putting in the blog.
Kunaal Naik: Namaste Alex, Welcome to the CLL podcast. How have you been?
Alex Weidauer: Namaste Kunaal, I am super well. Thanks, Kunaal, I’m excited to be here with you and chat about Rasa open source and the community.
Let’s get to know more about Alex’s daily routines, how rigorously he follows them, and how does he give himself a break between those hectic days?
It depends upon where he is. When he is in San Francisco, his day usually starts very early, because of the time difference (Berlin, Germany is 9 hours ahead of San Francisco, CA, USA) He wakes up early and goes for a run to free his mind. When he is in Berlin he sticks to the habit of going for a run, meditating, and then he spends some time reading non-work-related things. During weekends, he gets out of the cities and goes hiking or does something in nature to switch off a little. Running is his morning routine from the last 5 years and carries running shoes wherever he travels!
Why did they close the company and how did they learn “How not to make a product”, and how did it leads to the formation of Rasa?
The things mentioned above are just highlights of Treev. But the story of Treev for them includes just a lot of things that didn’t work. They realized that something that can sound like a good idea, in theory, may not necessarily mean that people use it and solve them as some people say hair on fire problem. One Product market key piece for them was retention, meaning users should not only be excited about the product, but they must keep using it — otherwise, it is hard to monetize. Something which never happened for Treev. Of their total 5000 users, around 10 people were using it every day which was not a good retention rate.
For a few people, it worked well but for most others, it didn’t, and hence they decided to shut down Treev as a product. The idea of Treev as a search bar was to turn it into some form of a system that you can chat with and ask questions, for example, “Where did I put the pitch deck of Rasa?” or something like that. But of course, it didn’t happen. At the same time, Slack opened its API for developers to build chatbots and that was the starting point for their journey into conversational AI.
They built their first bot using various bot frameworks like wit.ai and Dialogflow etc. It didn’t work for them because of the limitations those platforms have. That’s when they started building their in-house tool, which later evolved into Rasa open source-like framework, and here they are today with 3M+ downloads, 10K+ members on Forum, and 450+ contributors. Growth is absolutely phenomenal.
Talking about product-market fit. Tracking the usage of the developing product is one of the important parts. How is it happening with Rasa?
One of the challenges of building an open-source framework like Rasa is that there’s no tracking whatsoever. That’s something that also the community appreciates a lot. But some are stabbing in the dark, in many ways. But of course, there are other ways to look at the product-market fit.
The way they think about product-market fit nowadays is magical despite moments. Alex mentioned these ‘Despite Moments’ as early signals which indicate that you are on the right track to product-market fit. In detail, it can mean,
- Users hounding for early access to a closed beta.
- Even if there is no payment page, users are ready to pay.
- Users complain about your product but keep using it.
- Users spend a lot of time fixing the bug and/or reporting it.
- Businesses are ready to migrate to your product even if it is painful.
Those “magical despite moments” were there from the earliest days when Rasa NLU was launched in December 2016. Even before Rasa NLU was announced and launched, they already had somebody contributing to it by making a pull request on GitHub. Of course, they didn’t build the company on just that one moment so there were more moments like that for Rasa and not many for Treev.
Who were the first few members to join and what were they supposed to add to the company?
It was all very organic. The earlier version of Rasa used to provide conversational AI service to a lot of big companies. Back then Tom Bocklisch helped part-time while still working for his own company and then over time started working full time as Director of engineering. Rasa was also looking for a Project management role. That’s when Alex approached his friend Philipp Wolf who just finished his master’s degree. Today he heads Business Development full time. Few more folks from their circle join the company.
After a point, they were looking for a developer advocate who can dedicate their full time to the community and help those contributors contribute more, along with creating more educational content for developers. That’s where they found Justina Petraitytė who has been an active contributor. She had created a 2-hour Zero to Hero video on building a Rasa-based assistant which crossed 1M views on YouTube. She played an important role in building an engaging Rasa community that is growing very fast globally. Justina now heads and handles the Developer Relation team at Rasa. The rest of the folks who joined Rasa are as passionate and talented as the initial hire.
What were the biggest challenges the Conversational AI space was facing back then and now? What do you expect from other open-source NLP/AI communities(spaCy or Hugging Face🤗) to uplift the overall chatbot ecosystem?
Since they launched Rasa NLU in Dec 2016, there has been tremendous research taking place in conversational AI in the last few years. NLP research is going especially strong. Back then, the realization for Rasa was, that even though NLP is important for conversational AI but it’s not just about NLP, as it also involves handling and managing complex and complicated dialogues. Rasa is also involved in bridging the gap between the latest and greatest research and production and what developers can use.
Additionally, they are also involved in doing research that led to Rasa Core and the DIET Classifier. Rasa is always cutting edge and allows developers to use state-of-the-art components, like BERT for example. Recently Open AI has done an amazing job on the language generation part using GPT-3. Also, the library level like SpaCy, Transformers from Hugging Face, Flair, etc. are innovating one area, where Rasa is innovating dialogue management and conversational AI in general.
Rasa NLU got immediate traction on TechCrunch in Dec 2016. What is the biggest learning from early adopters that is now a big part of Rasa?
In earlier days, Rasa received an email from one of the largest banks in the world mentioning that they have deployed Rasa NLU in production. They regularly saw that the large companies are interested in an open-source solution but at the same time, they struggle to operationalize it. This means tools are needed to not just explore but create the whole process around it. In the case of Rasa, it is building an AI assistant with a team. This is something that influenced Rasa’s product decision and development. Specifically, Rasa X which is the tooling layer on top of Rasa open source is used by teams in many companies having workflows around that topic.
During the earlier days, there were lots of open-source projects for those workflows. For example, labeling training data which ideally involves Excel spreadsheets today as well. So large companies, want these workflows to be synced up in a nice way, which doesn’t involve manual work and Excel spreadsheets. As Alex mentioned earlier, conversational AI is not just about NLP. Along with NLP, there is dialogue management and language generation, and it is also all about workflows. This is something Alan mentioned about Conversation-Driven Development during Rasa developer conference L3-AI. The idea is really that you start with like an early prototype of an assistant and then, through real conversations, improve it, and learn from that. Something that still a lot of teams are struggling with. So Rasa is not only working on some of the latest research but also looking at pushing the boundaries in some of the latest barriers that industries are currently facing in terms of chatbots.
Rasa flushed service-based business and started focusing on building Rasa as a product with a strong focus on community-driven contribution. How do the company and the community benefit from this model?
Rasa wouldn’t have pulled off what it has, and be where it is today if they had focused on a service business. For example, the Rasa masterclass is a huge project they started in December 2019. And now it’s almost 100,000+ views on YouTube. Also, one of the important parts, machine learning research, would have been harder to pull off if they had focused on the day to day and services projects to stay bootstrapped. That was the major reason they decided to follow the Venture capital route, to fulfill Rasa’s mission to enable all makers to build AI assistants with a bigger and much better community, product, and education. Rasa raised $1.1m Seed funding from Basis Set Ventures followed by $13m Series A from Accel who also invested in Facebook, Flipkart, Slack, Spotify, and others in the past. Recently, Rasa secured a $26m Series B from Andreessen Horowitz(a16z) who are named super VCs invested in Facebook, Pinterest, Slack, Skype, GitHub, Flutter, Instagram, and others in the past.
Talking about education and community, what are the gaps that Rasa thinks the community can fill in all of these areas?
Rasa announced the Contributor Program in April 2019. The intention behind this was to channel the contribution and create a space where people with different backgrounds and skills can find ways to get involved. The program is to recognize their efforts and achievements within the community. You can get all the information in this blog post and from the podcast.
- Dec 2016 — Oct 2017 [7 months]: 30,000+ downloads, 300+ members, 40+ Contributors
- Oct 2017 — Apr 2018 [7 months]: 100k+ downloads[~233% up], 1,000+ members[~233% up] on Gitter and 100+ contributors[~150% up]
- Apr 2018 — Apr 2019 [12 months]: 500k+ downloads[~400% up], 3,500+ members[~250% up] on Forum and 300+ contributors[~200% up]
- Apr 2019 — July 2020 [14 months]: 3M+ downloads[~500% up], 10K+ members[~186% up] on Forum and 450+ contributors[~50% up]
How did Rasa handle the pace at which the community was growing? Was it overwhelming? Was this the time Rasa hired more employees?
It was a wild time for Rasa in many ways because of so much traction and pull from the community. At the same time, Rasa raised funding and with that money, they hired people from the product and community building side. But the whole intention was to keep up with the pace and make sure the community is happy and resolve open issues as early as possible. Additionally, they focus on building more tools around Rasa open source to support workflow.
What goes on behind procuring the funding and how much of your bandwidth is occupied to bring together all the aspects needed to ensure the procurement of the funding. What were your major learnings?
There is no short answer for this as discussing this would need another podcast. But in a nutshell, If we “Build a product that people want” then it makes the entire process smoother. In the case of Rasa, they have really strong traction from the community side and adoption of the tool from the business side. In 2018, the Rasa enterprise version fuelled the growth and attracted more investors. Also, In early 2019 investors were looking to invest in AI startups, and considering all the reasons mentioned above Rasa was a good fit.
RASA secured Series B of $26M led by Andreessen Horowitz in June 2020. What would these new investments and investors bring on to the table for Rasa?
In addition to money which they will mostly deploy on research and development. They just opened a research hub in Edinburgh and hired Adam Lopez from the University of Edinburgh. Also investing more in community education. They decided to work with Andreessen Horowitz to follow Rasa’s mission and focus more on the community and educating developers around the world. One more reason for working with them is because of Martin Casado, who co-founded and is a CTO at Nicira, which was acquired by VMware for $1.26 billion in 2012, and then he joined Andreessen Horowitz as a general partner. He has experience building many open-source companies including Open AI and that will help Rasa to build an even better company.
How did Rasa quickly adapt to going from a non-virtual event which was almost going to get canceled because of the Covid-19 to a bigger and better virtual event?
Covid-19 is an unprecedented time. But credit goes to the Marketing team who did a phenomenal job at managing this shift and doing things even better than everyone could have ever hoped for. Before that, in September 2019 they had their first-ever Rasa Developer summit in San Francisco which was attended by 250 people and was house full. During L3-AI their original idea was to do the conference at different places at the same time. But Covid-19 started to hit hence they thought of doing the same online which is better because even more amazing speakers and attendees can attend without worrying about travel. L3-AI was a huge success as around 5000+ folks attended with amazing speakers from all over the world!
Let’s have some fun moments with Alex,
Can chatbots evolve to a place where we ask for some business insights and it produces some of them by doing the analytics?
This is something they have talked a lot about — the five levels of AI assistance which they recently also updated. So, the higher we get in the levels, the less you as a user have to adapt to the interface and it does complicated tasks at some point in time. This is something that can’t be done by Rasa or Conversational AI alone. But it will be possible when we connect AI assistants to other systems to make it work flawlessly. For example, in the case of Looker, the BI tool user can get insights about “How is the business going?” or something because he understands what important metrics are for your business, but at the same time how to use the UI of Looker to get that information. When conversational AI is integrated into the tool like this then it will allow any user to become an expert for any piece of software, and without knowing how to use a tool we think that’s exciting, but it also would mean that it has to work with other tools and systems.
How did you come up with the name Rasa, and what is the current focus in terms of the future?
The internal name for this was a Parsa, Parsing text. When they were thinking to open source they felt like Parsa is kind of a weird name, and so they were looking for alternatives. They ended up seeing that Rasa.ai was available as a domain. And, that just sounded like a cool name but in the context of machine learning it kind of makes sense because “Tabula Rasa” in Latin means like a blank slate. So in machine learning, it all depends also on your training data. And, so that was something that just felt like it makes a little sense. Also in a bunch of different Indian languages, it always means something really good and even King is one of them.
In terms of focus, it has always been investing and spending a lot of time on product development and research. With recent funding, they will be doing that even more now. And then the other big piece is educating developers on building conversational AI. It is a long journey, and they are excited to see that the community is helping a lot with that. And it’s just super cool to see all the amazing efforts that have been going on. Just to name one, of course, is the Co-learning lounge on YouTube where this podcast will be posted, and it’s just super cool to see that the Co-learning Lounge has been doing such a phenomenal job there.
Do you watch any Indian Movies? If yes, then who is your favorite actor?
If you like what we do and want to know more about our community 👥 then please consider sharing, following, and joining it. It is completely FREE.
Also, don’t forget to show your love ❤️ by clapping 👏 for this article and let us know your views 💬 in the comment.
Join here: https://blogs.colearninglounge.com/join-us