Chat GPT has triggered an AI boom that has led to a multitude of new applications and discourses in business and society relating to artificial intelligence. As a provider of highly specialised cloud services for companies, it is crucial for us to show and provide customers with concrete ways in which they can use technological trends such as AI to their advantage. But how do you manage to be 'state of the art' in this fast-moving industry, i.e. always up to date and at the forefront? At CONCETO, we succeed by attracting new talent who enrich our opportunities with fresh ideas and up-to-date knowledge from universities, start-ups or other environments. One such example is our colleague Anny, who has a Master's degree in computer science specialising in AI from the University of Bonn, one of 10 German universities of excellence. She is significantly involved in the development and implementation of the latest solutions with a focus on AI. Today she took the time to answer our questions.
Hello Anny, please start by telling us about your role at CONCETO and what it entails.
I am a developer for our web applications and currently lead the technical project management of our AI Extractor. So I am the technical side of the project. There are also other sides, such as marketing, with whom I regularly discuss how the project is developing, what requirements there are and how these can be implemented. Such interdisciplinary meetings are very helpful when evaluating a new feature. We clarify issues ranging from customer requirements to user guidance. Together, we don't lose sight of the big picture and end up with a solution that really helps the customer.
So you're currently working on the AI Extractor. What exactly is this software?
Let's start with the problem: Many companies still send their invoices, delivery notes or other business documents as PDFs. This causes a lot of work for the recipient and is a source of errors, because someone spends the whole day manually transferring the data from the PDFs to the ERP. This is where we come in at this non-digitised point. The AI extractor reads the relevant data, checks it and sends it. The AI shows its full strength when the PDFs are unstructured, which occurs due to different formats and individual company designs. The total amount can sometimes be at the bottom right, sometimes at the top right or only on page 3 - the AI recognises the relevant values and formats them into standardised artefacts that are automatically fed into the ERP. In this context, you have probably heard of OCR, or Optical Character Recognition. This process has been around for a long time and works quite reliably, but is limited to character recognition. For example, an OCR can read the value of €5.30 on an invoice, but whether this is the net amount or the gross amount is beyond its capabilities. And this is where we come in with AI tools.
Were there any particular challenges in developing the AI Extractor?
This project was the first time we used artificial intelligence in a solution for our customers. Although we had already worked with AI before, it presented us with new challenges in addition to normal project management. AI is currently a hype topic and much is still under development. This puts a lot of pressure on the market, as interest is enormous, expectations are high and the competition is not small. However, we don't want to chase the trend or raise unrealistic expectations, but rather build optimal solutions with the most innovative technologies.
An important aspect of this is the safe handling of AI. Word has got around, but many private individuals and customers are still unaware that large language models such as ChatGPT can lie. This is called hallucinating. The programme does not think like a human and cannot understand things logically, but says to itself: "If I were a human, what would be the most likely answer after this sentence". This usually works really well and it's especially great for coding because code is also a language. The most probable next sentence is just logically not always the right one. An answer to the simple question of how many "m" there are in the word "always" can produce an incorrect result because the programme cannot actually count or logically process the letters. These are challenges that you naturally have to face and that you should always keep in mind.
Although we don't primarily work with LLMs, it is essential to handle every AI model responsibly. Nobody should be blinded by the hype and unrealistic promises. As the technical project manager, I am proud of how confidently the team overcame the hurdles of the project and how successfully the foray into the field of AI at business level got off to a good start.
"What research shows again and again is that every finding has a value. (...) This is in the nature of experimentation, because you never know exactly what these experiences can be useful for."
What does the use of AI look like in your and your colleagues' day-to-day work and what impact does it have?
We use AI to support programming. It's not what some people think or what is sometimes in the news: programmers will not become redundant and not everything can be written by ChatGPT. And that's a good thing, because you can't - as I mentioned earlier with hallucinations - completely rely on the output of artificial intelligence.
But AI is a great help, especially for simpler tasks, because it simply saves time. Of course, I can code a loop myself, but it's much faster this way. Nevertheless, the result is always checked to ensure functionality. In addition to coding, they also help with brainstorming or when you are looking for suggestions on how to implement something. The generated suggestions provide a good basis for further development.
It can be dangerous to use these programmes unsupervised. This is especially true for people starting out in their careers, because they can learn things that are not correct or learn nothing at all and let ChatGPT write everything. However, we want them to use such tools to support them because it is important for their day-to-day work. That's why we make sure to teach basic skills so that they understand what exactly happens in the code. This allows the result to be optimised afterwards and leads to a better product. At the same time, we teach the responsible use of such ubiquitous tools.
To what extent did the research-based approach from your Master's programme influence you?
In academia, you work with bleeding edge technology and, in the best case, you are involved in its development yourself. These are technologies that have only recently been published and for which there is usually only one paper, but nothing has been implemented yet, there is no library, let alone anything on the market. And that challenges you to develop the ideas yourself and always keep up with the latest technology. This combines well with what companies need: constant improvement, stability and future-proofing.
What research shows time and again is that every insight has a value. If I tackle a problem and try out five solutions and none of them work, this is initially a setback. But it is still a realisation because I know that I don't need to pursue these five approaches any further here; these findings can be a valuable approach for other applications. That's the nature of experimentation, because you never know exactly what these experiences can be useful for. Through research I have also learnt that there is actually always a workaround for problems. There are only a few things that cannot be solved. You might have to go in a different direction and get creative, but somewhere along the line there is a path that leads to the goal.
Apart from artificial intelligence, are there any software trends or developments that you think will have a significant impact on the way companies operate in the coming years?
We will see how trends continue, even if the hype surrounding AI will gradually stabilise. At the moment, we are mainly seeing large language models, but I believe that it won't just be generative technologies that are pushed in the future. For example, everyone knows Amazon's "You might also like these things" - that's also AI, an area called machine learning. It's been around for a while, it's just not yet as effective on a mass scale. However, development in this area is of course not standing still and I believe that we will see a few more things in this direction.
This area, which is very interesting for us, includes unsupervised learning, for example. This allows insightful information to be extracted from large volumes of unstructured data. A simple example: Let's take my personal calendar as a data basis. As a viewer, I can see the appointments, but machine learning can recognise the underlying relationships. For example, it can tell me how likely it is that I will go on holiday in the next quarter, how long this holiday will last and where it will probably take me. If you add other values, such as weather data, the predictions become even more precise. It is these interconnections, which humans can only recognise with great effort - if at all - where machine learning comes into its own.
Finally, could you give us an insight into a current project or a new idea that you are currently working on or that you are planning for the future?
At the moment, I am fully occupied with the AI Extractor. In the long term, it opens up many possibilities that we can and want to utilise. For example, what I just mentioned with machine learning: If invoices, orders or any other documents are sent to the AI Extractor, it extracts the data and forwards it to the ERP. As an intermediate step, this data could be used for business intelligence. If you know which customer buys what, when and at what price, you can draw well-founded conclusions that provide significant support for business decisions. This is, for example, a direction in which we would like to move.
Thank you for taking the time to give us an insight into the innovative side of CONCETO. We wish you continued success and fun developing new solutions!
Compatibility of family and career
A lot has changed for Matthias both privately and professionally since he started at CONCETO. In this interview, he explains how he manages to fulfil his various interests in everyday life.
Remote Work Down Under
Nico has broken new ground for CONCETO: He spent a year travelling around Australia by car with his girlfriend and continued to work part-time from there. We asked him about his experiences.
From the working student
to the Consultant
During his studies, Sebastian started working for us as a student trainee and wrote his bachelor's thesis here. We asked him to tell us more about his experiences as a working student.
Become part of the team
Then please send your complete application documents to Mrs Nina Vanschoenwinkel.
If you have any questions, simply contact us at +49 (228) 53 40 97-0 or send us an e-mail.
