Artificial intelligence (AI) and machine learning are not new topics, but have been researched and developed for many years. So far, however, they have tended to be niche areas, with only a few breakthroughs and findings.

This changed abruptly with the launch of ChatGPT. Everyone is talking about AI now, especially the field of “GenAI” – Generative Artificial Intelligence. This makes it possible to create media such as texts, images or videos using natural language input.

This breakthrough has captured the imagination of many people. We at SOPTIM also believe that AI harbours considerable potential for our SOPTIM Elements platform.

And that was how the AI Discovery Days came about: 

As part of the AI Discovery Days at SOPTIM, developers and product owners were brought together in a project to engage practically with the topic of AI and build up knowledge. To this end, an external consulting firm first introduced the topic of “AI in AWS”. AWS already offers a number of services for working with AI.

Two use cases were defined for the specific project content, each of which was then worked on by a project team within a week. One use case focused on classic machine learning, the other on GenAI, in order to cover a broad spectrum.

For the machine learning approach, the field of time series analysis, and in particular the topic of time series forecasting, suggested itself. Amazon Forecast offers a corresponding AWS service for this, and in its tried-and-tested SOPTIM Plus Forecast, SOPTIM also offers an optimal benchmark for assessing the quality of results.

Although the service could be set up quickly and easily, the quality of the forecast was clearly inferior to the SOPTIM Plus forecast. This was not surprising, as SOPTIM Plus Forecast sets impressive standards as a high-performance solution that has been optimised over many years. What was surprising, however, were the limitations of the service, such as the long training times and the restriction to 500 data points in the future, were surprising.

This use case aimed to implement a chat bot that can answer questions about SOPTIM products and interact with the ticket system. Firstly, separate documentation was created for the AI, using a knowledge base. Within a day, the team had developed a chat bot that was able to generate answers from the internal documentation.

Next, the AI was enabled to create and comment on tickets on a Jira board. This was done using so-called “agents”, which require an interface specification. The AI then decides independently when it needs to call up the corresponding endpoint from the specification for a request.

The remainder of the project was used to make more files available to the AI. This is where the first pitfalls occur, as the quality of the data is essential. If current information differs from historical information, the AI has problems generating the correct answer. The structure of the data also plays an essential role. The greater the granularity with which the data is stored, the more precisely the AI can refer to the corresponding document and thus identify where the information originated.

In an impressive result, the chat bot was quickly set up and put into operation. However, in order to create a reliable tool with real added value, further investment is required in the cleansing and structuring of the data.

AI, especially GenAI, remains an exciting topic. Working with AWS and the services provided by Amazon makes it easier to get started with AI. Nevertheless, investments in data quality and error analysis are necessary in order to achieve reliable results.

In the future, we will continue to pursue the topic and look for suitable use cases to integrate AI into our SOPTIM Elements platform, prioritising the consent and needs of our customers.