LANZ.AI: High quality training data for machine learning
Machine learning and artificial intelligence are becoming increasingly important in today's world. No wonder that the need for training data is also growing. They are needed to 'feed' artificial intelligence with the necessary prior knowledge and enable machine learning. With the help of LANZ.AI, companies can outsource this creation of training data - and thus concentrate on their actual work around artificial intelligence. In an interview with 5-HT, Daniel Lanz talks about an area of artificial intelligence that has so far received little attention - and which his startup has now taken on.
LANZ.AI is: Enabler and Accelerator
"LANZ.AI works with the impact sourcing company Digital Divide Data (DDD), which employs more than 500 people for the input of training data that we can access very quickly," explains Daniel Lanz, CEO and founder of the LANZ.AI start-up, "Our customers provide the raw data. This could be, for example, 100,000 camera images from the quality control area. In addition, the customers give us an exact description of how the product should look and how defects can be detected on the images. From this we create the specification for the so-called tagging or labeling of the images and discuss with our customers in great detail whether we have taken everything into account. This is the most important phase, because now it's all about quality. The procedure is then applied in a pilot phase and the results are checked, so that we are sure that the customer gets the data he really needs. This thorough preparation often accounts for a third of the total project duration. When everything is clear, we annotate the images. For example, we check whether there are any defects in the camera images of the products and mark them on the images. LANZ.AI thus identifies and marks objects and patterns in image data to create high-quality training data for machine learning.
For our customers, the creation of high-quality training data would be a very time-consuming task. We enable and accelerate the implementation of their solutions by doing this work for them. LANZ.AI is therefore Enabler and Accelerator".
Up to now LANZ.AI consists of one permanent employee, namely managing director and founder Daniel Lanz. However, this is by no means a disadvantage, as can be seen from Daniel's professional career.
LANZ.AI is: Over 20 years of experience
"I have worked for over 20 years in the fields of digitisation, data acquisition and software automation and was for many years the managing director of a digitisation company. That's why I know how great the need for high-quality data for process automation is, and why specification and a reliable partner for data entry are so important," says Daniel, reporting on his own experiences as managing director of the software and digitization company Content Conversion Specialists. But although training data is so important for process optimization, it is often neglected. Nobody here wants to spend five hours annotating photos. That's why training data is often created more poorly than right by project managers or working students. The results of the AI solution are correspondingly poor".
For an optimal result you need the best possible set of training data. In addition, machine learning is becoming more and more important in more areas, as the LANZ.AI homepage also reveals: from the identification of road users when driving autonomously to the early detection of illnesses in the health sector and the recording of handwritten documents in administration or in the field of cultural heritage.
"For the quality of the training data, the detailed elaboration of the specification is essential. This is often lacking for service providers who also produce training data. Since I know this problem from years of project experience, I finally concentrated on this topic and founded LANZ.AI in 2018", Daniel explains, "I find the topic of artificial intelligence and training data very exciting. It is an interesting future area in which I still see a lot of potential. Because it offers new possibilities for dramatic improvements in so many areas. The fascination for automation and process improvement is the reason why I founded LANZ.AI."
LANZ.AI is: Social and trustworthy
This is also reflected in the partners LANZ.AI works with: "We work with Digital Divide Data, a nonprofit organization whose goal is to bring young, underprivileged people in Africa and Southeast Asia into employment and thereby open up sustainable opportunities for them.
When it comes to outsourcing, people often think: People are being exploited or jobs are being lost in Germany. With DDD's impact sourcing approach, on the other hand, we ensure that people in Kenya, Laos and Cambodia have real prospects. These people are very committed and motivated to work on data entry jobs that are rather listless and ultimately of poor quality in Germany. And the project staff in Germany can take care of higher quality jobs that they probably enjoy more.
I was myself on site with DDD in Laos and have worked with DDD for more than 10 years in many large digitisation projects".
Daniel emphasizes that it is important that the concept of social entrepreneurship and impact sourcing is used to tackle social problems in the countries in an innovative way.
LANZ.AI is: Independent of industry
"Sometimes, however, confidential data must remain in Europe or even within the company building in Germany", Daniel explains, "This is also no problem for LANZ.AI. We also work together with German-speaking partners in Eastern Europe and can realize relatively many scenarios".
"Our first customers are from the agricultural and aerospace industries. Satellite imagery enables a much more efficient management of resources in the agricultural sector. For example, it is possible to determine which fields need to be irrigated or can be harvested, or in which areas there is pest infestation. But training data and machine learning can also be helpful in the maintenance area of large chemical plants, for example in the automated inspection of pipe systems. People get tired and make mistakes when they watch hours of video recordings. Computers provide better results in such cases".
However, there is no restriction on the areas in which training data can be edited by his start-up: "LANZ.AI is not industry-specific. The only thing we do not do is the so-called specialist tagging." This refers to the processing of training data by experts in the respective field. At first glance, this seems to be a limitation, but, as Daniel explains: "In surprisingly many cases, no special domain knowledge is required for tagging. All you need is a precise, unambiguous definition for the possible labels and tags. Images, for example, are even language independent. We can therefore also work very well internationally".
LANZ.AI is: Brave
In comparison to other countries, Daniel sees some difficulties for start-ups in Germany: "I think that in the USA, there is very positive and optimistic cooperation with start-ups, especially with regard to risks and opportunities. In Germany, on the other hand, it is a major shortcoming that people are hardly encouraged to take this step. It is therefore often difficult for founders. Of course, it is necessary to take a critical look, but encouragement is particularly important when setting up a company".
This could also be related to the so-called 'error culture', which he criticises, particularly in Germany: "In Germany, unfortunately, the focus of errors is on the damage caused by the error and not the learning effect. But without mistakes you cannot make progress. Mistakes in themselves are not bad, because it's through them that I learn the most."
He therefore recommends that other start-up entrepreneurs also see mistakes as opportunities for improvement. "I would also like to see more courage in Germany, especially when it comes to start-ups. Networks can make a big contribution here. They can not only establish targeted contacts, but also encourage founders who are often confronted with doubts. I experienced the same thing with 5-HT," recalls Daniel.
From start-ups in the 5-HT network, Daniel hopes for new partnerships for LANZ.AI in the area of machine learning and training data. Companies that want to outsource the area of training data or would like advice on this are of course welcome at LANZ.AI.
Become part of the 5-HT Digital Hub Chemistry & Health
Exchange ideas with innovative startups and future-oriented companies in our ecosystem. We look forward to meeting you!