LexaTexer’s Günther Hoffmann Is Using Artificial Intelligence To Make a Difference

Great strides have been made in the field of artificial intelligence over the past few years, leading to the achievement of feats that were once science fiction.

Today, artificial intelligence can predict the shows you might like to watch. It can help vehicles to navigate their environment without a human driver. More recently, it has become capable of generating content in a variety of formats, spanning images, videos, audio, prose, and even poetry.

These advancements have captivated public interest for good reason, but AI hasn't always enjoyed the current level of hype.

Günther Hoffmann knows this best. As a Berlin-based researcher and entrepreneur with a PhD in computer science, he spent years conducting research on AI and machine learning before the field gained significant traction.

His experience in academia and a stint in the United States fostered his passion for forging meaningful connections between academic research and real-world applications. That passion transpired into two startups he founded: LexaTexer and MedaPlus. LexaTexer uses AI to automate the development of data-driven solutions, and MedaPlus is an AI-assisted software platform that can deliver predictive diagnostics for respiratory and heart signals.

We interviewed Hoffmann to uncover the latest trends in artificial intelligence, the progress he has made with LexaTexer and sister startup MedaPlus, and his future plans for these ventures.

EM: LexaTexer and MedaPlus are among the top 10 startups of the InnovateSingapore programme for the retail and healthcare sectors, and MedaPlus has been shortlisted as one of the top startups for SLINGSHOT 2020. Congratulations on the achievements, and share with us what differentiates what you do from your competitors.

GH: Thank you. A lot of trial and error is required when creating new products. In domains like AI, there is a clear need to differentiate what could be done in the future from what can be done today. Listening closely and in some cases, co-innovating with our customers, have helped us to create products that impact real lives today.

For example, we co-innovated our product LXTXR Smart Power with Salzburg AG, using running data from over 30 power plants to generate an AI-based solution that supports utilities by extending the remaining useful life of their turbines. The same system is now used by facility managers to increase the efficiency of their buildings.

With MedaPlus, we have done similarly by working with doctors from Germany and Switzerland to ensure that the data we use for AI-driven analytics is trustworthy and high in quality. We even went the extra mile by producing a system capable of explaining its findings, which is sometimes termed as explainable AI.

The ubiquity of AI has brought about the need to understand how AI systems work and how they make the decisions that they do. Explainable AI is the framework to help humans interpret the journey of data from the time of input to the final predictive output made by machines.

Explainability is critical as the industry leans increasing towards deep-learning and neural network models for problem solving. Neural network models are difficult to comprehend and at times undecipherable. They are also fallible and can create biases. If deployed for critical use cases, incorrect decisions by these ‘black box’ models can cause substantial negative impact.

Use cases with significant impact on humans, such as those in the military or healthcare, would therefore fare well by using algorithms that are interpretable. Additional effort is needed to achieve the required accuracy, but it is important to be transparent about the decision-making processes machines use for high-impact use cases, notwithstanding the need to comply with regulatory requirements.

This also strengthens confidence in AI systems, safeguards against biases and helps developers address vulnerabilities. For use cases where impact is less profound, like AI-based chatbots or sentiment analyses of applications, concerns over the explainability of AI and how ‘black box’ models work are diminutive.

‍– Vinod Anand Bijlani, AI & IoT Practice Lead, Hewlett Packard Enterprise

EM: Why is artificial intelligence drawing more interest over the past few years?

GH: AI is not a new technology. It has been around for many years now and we have seen varying adoption rates in different industries. From what we observe, students who have learnt about AI and understand its potential applications have now taken on decision-making positions on the buyer’s side. This has helped to speed up adoption.

Countries like Germany, Singapore and Japan have to manage challenges such as their ageing populations, which has encouraged governments to co-innovate with technology practitioners more actively than before to address these issues.

Now more than ever, I anticipate AI to be a category of technologies that will have an increasingly significant impact on the way we organise our lives in the long run.

EM: What advice would you give to companies that have yet to go digital?

GH: Broadly speaking, we observe two different types of companies. One, companies that have started monitoring and collecting data many years ago, even when they did not have the means to analyse and gain immediate benefits from doing so, and second, companies that are still trying to understand what kind of use cases and benefits they could derive from going digital.

Companies in the former category may now find that they can benefit from the wealth of data they have previously gathered, and will have a head start.

In the context of LexaTexer, a tier-one automotive supplier we work with has collected data for a long time without a concrete application in mind. They have now benefited by putting those data to use through an application co-developed with us, which has increased their overall equipment efficiency (OEE) by 3 to 17% since.

Overall equipment effectiveness (OEE) is a compounded metric to monitor the availability, performance and quality of manufacturing operations. For example, if your manufacturing line has installed capacity to produce 100 pencils a day and ended up making only 50, you can use the metric to determine the root cause behind it—how much of it was driven by lower availability of machines, inferior performance or poor quality.

In the past, we would track OEE at periodic intervals and use a variety of production reports as inputs to compute the values on a daily or monthly basis. But now, thyssenkrupp has its own IoT platform to connect the machines in real-time and monitor OEE values as they happen, with the added benefit of making corrections and improvements immediately. We have also started deploying machine learning algorithms and edge analytics to optimise the production processes and predict equipment failure, all of which contributes to OEE improvements.

To put in perspective, even a 5% improvement in OEE can have one to three times the impact on the bottom line of industrial plants.

– Abhinav Singhal, Head of Strategy and Transformation, SAP (formerly from thyssenkrupp)

Another example is a producer of welding machines for large tankers and vessels. They amassed a lot of data from their welding robots but were unsure how to put that to use. We were able to step in and use those data to build applications that enable non-experienced welders to create complex structures with a welding torch, simply by routing the data through these torches.

Contrary to off-the-shelf software, AI solutions require numerous experiments to pull off, but for companies that are willing to accept that as part of the development process, they will often find success at the end of this process.

EM: What are your plans in Asia, and how does IoT Tribe’s Deeptech Accelerator programme fit into these plans?

GH: The countries in Southeast Asia are now leading growth in many industry sectors, and countries like Singapore have made the most of digitisation initiatives that are yet unseen in Europe. In some of the niche markets we are active in, like hydropower, Southeast Asia is by far the largest and fastest-growing market. New technology is also rapidly adopted in Asia. IoT Tribe’s Deeptech Accelerator helped us to identify the right approach to target these markets, and connected us with their network of specialists. We appreciate IoT Tribe’s help to get us up and running in the region.

The investment outlook for early-stage deep technology startups in Southeast Asia remains robust, driven by the importance of AI within Industry 4.0. Covid-19 has transformed the way people work, live, and learn, creating enormous opportunities for founders to create novel solutions that leverage “advanced analytics and data as digital oil”.

Over the next few years, there will be increased SPAC and IPO deals in Southeast Asia, through which founders and venture investors will use recycled capital to back the next generation of promising startups.

The strong growth of the startup ecosystem in Southeast Asia has been a positive factor for founders by democratising access to networks, mentors, talent, and capital. As the saying goes, it takes a village to nurture a startup, thus I will encourage founders to surround themselves with like-minded stakeholders.

– Eddie Ler, Partner, Vantage Venture

EM: What tips do you have for startups that want to emulate the achievements of LexaTexer and MedaPlus?

GH: I’m afraid there is no recipe or silver bullet. But persistence and a capability to learn from failure may certainly help.

EM: What are you passionate about besides AI and technology?

GH: I’m quite fond of skydiving and jumping out of perfectly intact airplanes.