• Privately I often have a few side projects in a year, in order to widen my tech stack or try new technologies that I do not have contact with in my business. In 2023 I was organizing a New Years Eve party with my wife and we were planning a murder mystery dinner. Frustrated by the lack of good options for such games, I decided to build my own with the help of GPT-4 and Dall-E-3.
• For testing, multiple games were played by several independent groups of testers, and the feedback was overall very positive. I decided to offer the generation process as a service and started to orchestrate the solution in python-based microservices.
• Because of my lack of frontend and UX related skills, this was done in collaboration with a web-marketing company from Cologne.

Interhyp is the largest residential mortgage provider in Germany and one of the major players in the financial industry.

• The company conducts a large part of its day-to-day business by telephone. As part of the technological renovation of the company, several historical systems are to be replaced by modern approaches. A monolithic application is gradually being translated into a microservice landscape and specialized on-prem servers for telephony are being replaced by cloud-based applications. 
• The aim of the project was therefore to enable the transition to the cloud-based telecommunications platform Twilio and to remove the dependencies on the monolith. The customizable Twilio Flex front end was used and micro-services were created to manage sensitive data in the private network.

• Blickfeld is one of the up-and-coming start-ups in the greater Munich area. Its 3D scanners (LiDAR) are in high demand in the smart city, automotive and security industries.
• After 3 years of development based on a Xilinx SoC with FPGA, it became clear that this approach had too many limitations and also held back further improvement. In addition, the overall project became more and more fragmented and poorly documented due to several research projects within the company. A new architecture was designed with a Raspberry Pi CM4, several STM32s for real-time applications and other improvements. This project was primarily about the migration of FPGA tasks to software, the optimization of code, and the automatic validation of results.

• A key criterion for a modern LiDAR device is its accuracy. That is why Blickfeld is constantly trying to improve its results by adapting models and experimenting with new, previously unknown approaches. This project also involved developing new approaches to improve scanning accuracy. This meant simulating the device?s behavior and validating it with real-world data. The result was algorithms that use reference data to correct the system during runtime.

• BayernOIL is one of the larger refineries in Germany, based in Neustadt, Bavaria. The refinery employs over 700 engineers who process crude oil into refined gasoline and other products.
• Like other older companies, BayernOIL struggles with documentation consistency. As the refinery was built several decades ago, many processes there are still manual, including over ten thousand manual valves spread across the facilities. BayernOIL wants to digitize these in order to integrate the information into their new digital workflows.

FormelD is a well-known and valued partner for automotive OEMs in Germany.Originally focused on documentation, the company now also takes on other complex tasks for the industry.

• FormelD aimed to digitize its material quality assurance centers at partner sites. This ambition led to the external development of software that would support the employees on site in their tasks and give the OEM immediate feedback if a supply chain had problems or was too faulty. 
• After an external partner had written the software, the result was found to be unsatisfactory. The reason for this was a high incidence of errors and crashes, as well as incorrect implementation of requirements. This led to this follow-up project, which involved improving the software and quality of an external

service provider.

• The Department of Computer Science at Nuremberg Institute of Technology has a broad spectrum of research work in the field of AI. The department wanted to develop a system that could apply modern speech recognition approaches and at the same time be easy to use for users without a technical background.
• The aim was to develop a scalable solution that allows existing language models to be easily combined with user-defined language models. These models are generated by providing documents to the system to dynamically generate the client?s custom model.

• Valmet Automotive?s development department is spread across several locations in Germany. These facilities develop specific parts for modern cars and also test them for local car manufacturers.
• Valmet Automotive Engineering had problems with audits due to insufficient documentation. They were also trying to stay competitive with other engineering companies that could do the job more efficiently than the company itself. After some analysis of the internal processes, several areas were identified where efficiency could be increased by reducing manual work and replacing it with digital methods.
• One of the biggest inefficiencies was identified in the area of driving tests, where many manual steps are required: from uploading measurement data to typing up subjective forms. A system should be developed to fully automate the process and thus allow more time for driving and fewer errors.
• Another area where potential for improvement was identified was the poor documentation of inventory both in the offices and in the workshops. A system was introduced that used an app to create a tree-like data structure of the inventory. The project has made it much easier to find tools and record certificates.