Daniel Beutel is the founder of Flower Labs. Founded in Hamburg, the company is a member of ARIC, Open Source works has made the leap: the leap across the pond to San Francisco as well as the leap to becoming a successful tech company. In this interview, Daniel talks about the founding journey, the community recipe for success and the potential of Federated learning.
ARIC: Who are you and what do you do?
Daniel J. Beutel: I am one of the three founders and CEO of Flower Labs and one of the developers of Flower’s Federated AI platform.
What exactly does Flower.AI do?
Flower is an open-source framework for training AI on distributed data. The approach is called federated learning or federated AI. Organizations such as Nvidia, Siemens and J.P. Morgan use this approach to train AI models on sensitive data that they were previously unable to use.
When did you realize that the topic of federated learning had potential?
We realized early on that the biggest challenge in AI projects is usually getting access to the right training data. Data is often legally protected, it either has to remain with the user or is spread across different company silos. And it is right to protect this data, because it is usually sensitive and very valuable data.
But if the data is not accessible, how can you train a good AI model? That was the trigger. When we came across federated learning, we realized that federated learning can actually solve exactly this problem, because with federated learning you simply leave the data where it already is. Nobody has to disclose sensitive data, but you can still work together and train AI models on this data.
How long have you been around?
The startup Flower Labs has been around since 2023, but the Flower open source framework has been around for longer, since the beginning of 2020.
Did you find the doors open or did you have to do some convincing?
When we published the open source framework back then, Federated AI was still a topic that was relatively unknown in the industry. We had to do a lot of educational work. For example, every year we organize the Flower AI Summit, the world’s largest conference in the field of Federated AI. The topic has now reached many people and we no longer have to talk about it so much because we are approached by companies that become aware of us through the Flower open source framework.
How did you start? And did you start in Hamburg or was it already international?
Yes, two of us three founders lived in Hamburg at the time. Although we started in Hamburg, we were actually international right from the start because we developed Flower in collaboration with our colleagues in Cambridge.
Why did you decide to go open source?
We want to make this new approach to developing AI accessible to more companies and organizations. Open source is the best way to achieve this. Before we started, Federated AI was something that only existed in proprietary systems behind closed doors at Google, Apple and Meta. With the Flower open-source framework, we have managed to democratize Federated AI and establish the world’s largest community in this field.
This community has created a strong ecosystem around the Flower platform, which in turn benefits others in the community. If someone in a research institution, a researcher at a university for example, develops a new approach to training an AI model using the Flower framework, then someone who works in an industrial company and is simply trying to solve a current problem there can simply take this approach and use it to solve their own problem.
That’s why we do so much community work to help this community grow and become more connected. The nice thing about a community like this is that everyone ultimately benefits from the others. You put something in and get a lot back and end up with a great ecosystem.
What exactly does community work mean for you? What do you do to bring people together?
For example, we have an open Slack channel through which anyone can easily get in touch with the community. And AI researchers or AI developers often introduce themselves there and post what they are working on. And that often provides the impetus for someone else to say: “Ah, interesting! I’m working on something similar.” Or: “I have a question about that.”
We also organize many events, such as the Flower AI Summit, the world’s largest Federated AI conference. The last Flower AI Summit was in London in March 2025. In September, we had the
…and then you managed to get relatively large funding from Y Combinator during the not-so-easy Covid period. How did you manage that?
When we initially developed the Flower framework as open source, the topic of federated AI was still relatively unknown. Through our work in the open source project, however, we were able to observe how companies were using Flower and Federated AI. Towards the end of 2022, we realized that the topic was gaining momentum in the industry. Large industrial companies such as Bosch were taking the topic very seriously. And that was the impetus for us to say, “Now is exactly the right time to found a startup that focuses solely on this topic“. We then decided to found Flower Labs, applied to Y Combinator and moved to San Francisco for three months.
Based on your experience, is there anything you would advise other founders from our community who are just starting out with their AI start-ups?
My recommendation is to always work on something you are passionate about. Founding a startup is a lot of work and not always easy. The important thing is to have a topic that you are really passionate about. If you have a topic that you believe in, it simply gives you the energy to go down this path. That is the most important thing for me. I could hardly imagine working on something that I don’t really believe in.
Would you like to say something else?
I believe we have a huge opportunity because a shift is currently happening. Most AI models that are developed today are still trained centrally. That means you have to collect all the data in one place and then train the AI model there. This means that everyone who participates has to disclose their data.
We are seeing the beginning of a new era where AI is no longer centralized, but decentralized. Where different organizations can come together. And these organizations can say: each of us has a certain amount of data, a certain amount of computing power and if we network together, we suddenly have much more computing power and much more data. And that ultimately results in better AI models. This development is only just beginning and is a huge opportunity for various sectors in which it has traditionally been difficult to train good AI models.
Which sectors does this apply to, for example?
For example, health and life sciences. The data in this area must be very well protected, and rightly so. The only problem in the health sector is that nobody has enough data in one place to train good AI models because this data is scattered across individual hospitals. We would all benefit as a society if we had really good medical AI models. This ability to network and work together in a decentralized way is what makes it possible to train good AI models in the first place. Other areas are finance, logistics and production.
Why do we have this shift right now?
In the past, for example five years ago, it was very difficult to use this decentralized training approach. Now we have the infrastructure and we also have more AI researchers and AI developers who are familiar with these approaches. We have open source frameworks like Flower that make this kind of collaboration very easy. An existing AI project that works in a completely centralized way can be trained in a decentralized way relatively easily with Flower. And this simplification makes decentralized AI training accessible to more organizations and industrial companies. In the meantime, Flower has become the industry standard for federated learning and federated AI – we have thousands of users in various industries who use the Flower platform productively to train AI models on distributed data.
Interview: Sabrina Pohlmann
With our interviews, we want to introduce you to different perspectives and players in the field of AI. The positions of our interview partners do not necessarily reflect the positions of the ARIC.
Further interviews:
