How MsDAISIE came about

Lisa Heß : Chris, Björn, first of all, congratulations on receiving funding from the BMFE.

Björn Marz: Thank you, considering where we come from as an SME and how we have developed, we are incredibly proud of the path we have taken.

Lisa Heß : Okay, we'll talk about the "how" later, but let's talk about the "what" first. What is MsDAISIE and how did you come up with the idea?

Björn Marz : It was actually a conversation with a teacher that gave me the idea. We worked a lot with ChatGPT last summer, me for the first time, Chris Ehrling for years. An acquaintance was trying to create teaching material with it and was frustrated because the output wasn't very good. I gave him a crash course in ChatGPT, how to assign roles, how to use prompt engineering, etc. etc. And after the conversation, I thought to myself, why don't we build something that does exactly this work for the teachers. I then called Chris. Just 36 hours had passed between that conversation and the first kickoff (July 6, 2023) for MsDAISIE - and then we got started.

Lisa Heß : Okay, how did you then proceed?

Björn Marz: First of all, we wanted a project that was as agile and customer-oriented as possible. We always wanted to pick up the target group of teachers and obtain feedback in consistent rapid prototyping. Parallel to our initial momentum, we carried out market research and were surprised to discover that there was hardly anything else on the market. And, above all, that our "friendly teachers" don't use any tools at all. Incidentally, at Didacta in February 2024, we were again amazed to discover that there are hardly any comparable applications and that we simply generate really good results with "MsD", as we affectionately call it.

Chris Ehrling: Our goal from the outset was to create a tool that would make it easier for teachers to organize themselves, create materials and plan lesson sequences. Unfortunately, 50 to 60-hour weeks are normal for many teachers and only 30% of their working time is spent actually teaching the pupils.

On the one hand, we wanted to do something about the high workload for teachers and, on the other hand, create space for the tasks that really count in the teaching profession.

Lisa Heß: Okay, MsDAISIE was born. How did it get its name?

Chris Ehrling: We use our Daisie framework and Björn and I actually went into the kickoff with a working title that just fit.

Lisa Heß : What can MsDAISIE actually do and what makes it so special?

Chris Ehrling: The special feature of MsDAISIE lies in a technology called Retrieval Augmented Generation. This optimises the output of an existing large language model so that it can draw on knowledge outside its training data sources before generating an answer. This means that the model does not have to rely on what it already "knows", but can draw on very specific information, understand it and integrate it into the answers. This significantly improves the quality and relevance of the answer, as the AI stays up to date with the latest information and doesn't just rely on previously stored knowledge. Simply put, we didn't want to outdo existing LLMs, but rather complement them in a meaningful way.

To this end, we have provided our model with targeted curricula, year-specific tasks we have created ourselves and other didactic information.

In addition, each teacher has the option of uploading their own materials such as worksheets, exams, texts, graphics and images to a personal database in order to completely individualise the output.

With a view to career starters and lateral entrants, we have also built in the functionality to automatically create lesson plans.

Lisa Heß: Why has FIDA been able to develop MsDAISIE so successfully?

Chris Ehrling: We are an interdisciplinary team with a strong university background that has been researching and developing services and products in the field of AI for over 5 years. From the very beginning, we had a very strong focus on machine processing of natural language, so all the new possibilities offered by LLMs naturally played into our hands. Despite having a double-digit number in the team, we don't have a single duplicate in the courses we have completed and have been able to incorporate pedagogical, psychological and linguistic expertise into MsDAISIE.

Lisa Heß : Björn, you mentioned that it is something special for FIDA, why?

Björn Marz: Well, we are not a start-up and not a classic product company. We come from a consulting background and in one fell swoop we have a wealth of use cases and ideas for products. Chris and his team have been researching the topic of AI for 5 years and all of a sudden the train started rolling. We often had discussions about the scope and benefits of our solution and were always able to point out: "The teachers are involved and want it, let's believe in the result". It only became critical when our perfectionism got the better of us and we ran the risk of tearing up our schedule. You have to be agile and stick to the definition of done and we're back on track.

Lisa Heß: Okay, now to the promotion. How did it come about and what does it mean?

Chris Ehrling: We were and are absolutely convinced of our idea, but of course we also took a certain business risk during development, so we quickly came up with the idea of rewarding ourselves a little for our thirst for research. Karsten Völzke oversaw the application process with an external service provider for our projects, and MsDAISIE received the funding approval from them.

Björn Marz: The funding may not be anything special for others. For us, it is a real breakthrough and we are very, very happy that we have managed to be recognized in this way. Of course, the money is also nice, but we are well financed by our main business. Ultimately, it's now a matter of finding partners for marketing and bringing MsDAISIE to customers.

Chris Ehrling: ...we have considered two options for this, because as a medium-sized company we cannot serve 1.2 million teachers directly B2C. We have the option of integrating MsDAISIE as a module into existing applications via partners or selling package solutions.

Lisa Heß: Thank you Chris and Björn for your time.

Gotha, April 01, 2024