MLReef: Shaping the Future of Collaborative Machine Learning

Year after year, machine learning demonstrates what it is capable of: distinguishing cats from dogs, beating humans at chess and Go, and solving tough computational problems such as determining how proteins fold.

Yet, until recently, most enterprises did not adopt machine learning: they either did not understand its business value or lacked the expertise to implement it. However, since ChatGPT, at the very least, the potential and impact of machine learning on almost every sector are much more apparent—although the effort to get businesses to adopt machine learning in their operations is not new.

MLReef was founded in 2017 by Camillo Pachmann to help businesses get started with machine learning, collaboratively develop models with several stakeholders, and bring these models faster into production to realize business value. 

Learn more about the future of collaborative machine learning from our interview with the founder and CEO, Camillo Pachmann: 

Why Did You Start MLReef?

When I got into machine learning in 2015/2016, I got really hooked: Training my first machine learning model had this wow effect, making things possible that were previously thought of as impossible. Machine learning was about to change many things fundamentally, and so I decided to get into it.

But first, I wanted to get my feet wet and founded an AI consulting boutique that handled several machine learning projects in the industrial sector, for example, around automating production lines or driving efficiency improvement in logistics. However, I soon realized how painful collaborating with external partners could be – the more stakeholders, the trickier – and started developing an open-source tool to make machine learning development easier and more collaborative. That’s how MLReef was born. 

How Does Collaborative Machine Learning Work?

MLReef is like a workbench where different users share different tools to develop machine learning models, especially when pre-processing data and preparing datasets for model training. MLReef offers a high degree of flexibility, so you can try whatever you want. 

While designing new machine learning algorithms is mainly a matter of research, having good quality data is what makes a difference for applied machine learning employed for solving practical problems. The foundation of machine learning is data-driven.

Therefore collaboration should not be limited to a team but span across different communities and involve not only tech people like data engineers but also domain experts, for example, medical experts when building a computer vision system to identify tumors. This can really help to create a high-quality dataset. 

There are still lots of practical questions for enterprises that want to implement machine learning: How easily can I access data? Do I have a budget? Can I play around? And does the business unit understand the impact? 

For years, the main bottleneck for machine learning adoption has been the enterprise culture, especially in Europe. Europe has just started catching up on digitization but hasn’t established a machine learning culture. Countries like the US or the UK have long understood that data is an asset that needs to be made available to be able to process it and derive value subsequently. 

This might now have changed with ChatGPT – the technology has been around for years, but now end users can see the value it brings – and the hype might lead to a shift in enterprise culture. Before, you could get fired for implementing machine learning; now, you could get fired for not implementing it. 

While large language models like GPT-3 require immense amounts of budget, training time, and top-notch talent, for most practical cases, these giant models will be consumed through an API. Decentralized model development is the way to go to finetune these large models or train much smaller models, which then solve very specific practical problems.

How Did You Evaluate Your Startup Idea?

I simply loved the idea of doing something with machine learning, and that’s why I started a consultancy. It’s hard to assess and time the adoption of a single technology, and with machine learning, the market adoption has been slower than expected in the last few years. But we knew machine learning was here to stay, and we eventually realized that collaboration was a major bottleneck for developing machine learning models – and that’s when we decided to build MLReef as a product to solve that bottleneck. We’re now following a simple SAAS model, charging a license fee for the platform and a small markup if you use our cloud infrastructure. 

What Advice Would You Give Fellow Deep Tech Founders?

Talk to potential customers as much as possible from the early days! I started doing that and stopped at some point to focus on our product and setting up the architecture. But as soon as you stop talking to customers, you start floating in space and might miss building a product your customers value. The market defines the product: between market pull and tech push, the market always wins.

Further Reading

The Best Open-Source MLOps Tools You Should Know – MLReef is among them, featured on the neptune.ai MLOps blog. 

The 2022 Market Guide Reports for Multipersona and Engineering Data Science and ML Platforms – An overview presented on the MLReef blog. 

MLReef: NÖ-KI-Startup kooperiert mit Rüstungskonzern Saab – German article about the collaboration between MLReef and Swedish defense giant Saab.

Künstliche Intelligenz entwickeln für Anfänger – German article about how MLReef is empowering more people to participate in the development and use of AI technology.