atacama blooms GmbH & Co. KG is a software company based in Bremen that combines expertise in computer science, health science and management, as well as linguistics, with a focus on modern technologies such as artificial intelligence. The company has more than 25 employees and develops intelligent solutions for service providers, payers, and software vendors in the healthcare sector.
The project at a glance
- Evaluation of various open-source large language models (LLMs) to achieve the best possible results
- Consultancy on available models and their hardware requirements to support decision-making when purchasing hardware
- Realization of a proof of concept (PoC) for a medical report generator within a matter of weeks. The medical reports generated for demonstration patients passed the plausibility checks performed by medical specialists.
Initial situation
Hospitals and care facilities are increasingly struggling with staff shortages and work overload. At the same time, increasing digitalization in the healthcare industry offers great potential for relieving medical staff through more efficient documentation processes. The ongoing introduction of digital patient data and documentation offers the possibility of using generative AI models to automate medical documentation, particularly when writing medical reports.
The primary concern of atacama blooms was therefore to develop a system that extracts the relevant information from the digital care plan and converts it into structured and professionally correct medical reports. The constant technological changes in this young field, as well as the large number of available open-source LLMs, which are preferred to proprietary models such as GPT for data protection reasons, present a particular challenge.
Solution
codecentric developed a simple dashboard interface that allowed the selection of demo patients, the option of providing additional information about the person being treated, and the creation of automated medical reports at the push of a button. All processing is performed by the LLM in the background in several steps. These reports precisely formulate anamnesis, epicrisis, and further recommendations based on the available patient information.
An important aspect of the project was prompt engineering. This was a particularly complex task due to the specific medical terminology. We continuously gathered feedback from medical professionals over the course of several weeks and improved the system accordingly.
Result
Together with atacama blooms, codecentric has created an AI assistant that can not only correctly reproduce relevant medical content, but also make well-founded treatment recommendations. The automated generation of medical reports significantly reduces the time spent on manual documentation, reducing the workload for specialists and other medical staff in their daily work. Both atacama blooms and the initial quality controls performed by medical specialists judged the quality of the generated medical reports to be good. The quality of the medical reports generated can be further improved in the future by continuing to improve the LLMs and through specialized training (fine-tuning) of the AI models. atacama blooms still has a lot planned for the future, such as the productive implementation in the first pilot hospitals. We look forward to being part of this development, too.
Sebastian Zebbities
Managing Director, atacama blooms GmbH & Co. KG
The cooperation with codecentric was characterized by professionalism and a partnership of equals right from the start. Together, we focused on relieving the workload of hospital medical staff, thereby freeing up more time for essential tasks: providing patients with the best possible care.
Sebastian Zebbities
Managing Director, atacama blooms GmbH & Co. KG
Any questions about the project?
Want to know more about the project? Are you interested in a similar solution for your company?
Service Lead GenAI
Service Lead GenAI
Further projects of codecentric AG
Find out about further successful projects that we have completed with our customers. Perhaps you will find inspiration for a use case in your company.
Daniel Ladischenski