This is now a classic post that I am actually only writing so that my ChatGPT chatbot can provide information about my references.

Here are some of the projects I have been involved in over the last few years. However, this is only a small selection. If you would like to know more, you can of course send me a message at any time.

Accompanying the ChatGPT experiment of Helvetia Insurance Switzerland – Switzerland’s first LLM chatbot

As a research assistant at the Lucerne University of Applied Sciences and Arts, I accompanied the ChatGPT experiment of Helvetia Switzerland. I established the contacts with OpenAI, accompanied the process and provided support with the analyses. Since the successful release of the first ChatGPT chatbot in Switzerland, I have been supporting the Helvetia Switzerland team with questions about usability, best practices and the further development of the LLM chatbot Clara. The LLM chatbot is constantly being further developed, is being given new functions and capabilities and is increasingly replacing the intent-based chatbot. Tests are also carried out time and again with regard to usability and tonality.

You can find more information about the launch of the Helvetia Experiment on my blog. You can also hear more information about the project in the podcast episode with Florian Nägele.

Strictly speaking, this LLM chatbot project is not my first project with Helvetia Switzerland. In 2020, I supported the team in defining their chatbot personality. In joint workshops, we worked out the best way for the Clara chatbot to communicate with customers. We then used this to develop guidelines for writing the chat dialogs.

Challenges

• A pioneering project with regard to LLM chatbots
• The right prompt engineering and preventing LLM breakouts
• Migration of an intent- and rule-based chatbot to an LLM chatbot or generative AI chatbot

Highlights

• Satisfied users and a better NPS compared to the old intent-based chatbot
• Good roadmap for the further migration of rule-based flows to the LLM chatbot

An LLM chatbot for the SBB pension fund

The SBB Pension Fund receives a large number of inquiries from new and existing customers every day. Many of these queries can be answered automatically. As part of a concept workshop, we developed the scope and the most important requirements for their first AI chatbot or LLM chatbot. I then compiled all the technical and content requirements and found a suitable technology provider for the company. As part of the implementation, in addition to the general management of the technical provider, I was responsible for the prompt engineering and the test phase incl. Optimizations accompanied. Microsoft Co-Pilot Studio was used to implement this chatbot and supplemented with the help of the Microsoft Bot Framework. SBB Pension Fund uses a variety of Microsoft services, so it made sense to use these for the chatbot as well.

Challenges

• Finding the right scope for the phased development of the LLM chatbot
• The use of the new Microsoft Co-Pilot Studio
• Very broad customer and new customer inquiries, sometimes also on sensitive topics

Highlights

• Prompt engineering that takes into account the conditions of the Microsoft Co-Pilot Studio and prevents the chatbot from breaking out on sensitive topics
• Different tonalities or personas for the LLM chatbot
• Project plan with different phases

The further development of Enercity’s intent-based chatbot into the LLM chatbot

The energy company from northern Germany already had an intent-based chatbot with some rule-based chatflows. As part of an optimization project, I suggested changing the existing chatbot into an LLM chatbot or a generative AI bot. As a result, I was able to create the concept for the migration of all intents and am accompanying the migration. There is no new technical provider, but the existing provider can be retained.

Challenges

• Migrating an intent-based chatbot that has grown over the years to an LLM chatbot
• Partly poor content sources and a confusing website that is difficult to use for an LLM

Highlights

• Definition of a precise migration roadmap
• Preparation of knowledge content for an LLM chatbot
• Precise and well thought-out prompt engineering

LLM chatbot for the B2B company GS1

GS1 is the company in Switzerland that creates the barcodes. Many people don’t know the association, even though they use its products almost every day. Customers have many questions when it comes to ordering or using barcodes. GS1 opted for a chatbot shortly after the release of ChatGPT. As a result, the new AI chatbot was implemented with the latest language models (LLMs) from the outset and without intents or predefined flows. My task was to sharpen the use case, define the project scope and select the right technology provider. I then supported the chatbot team with questions about usability and prompts.

Challenges

• Choosing the right technology provider
• A very broad target group with many different requests
• Some sources of knowledge that were difficult for LLMs to process

Highlights

• Increasing the usability of the chatbot
• Adjustments to the tonality of the LLM chatbot

POC and tendering and selection of providers at Raiffeisen Switzerland

I worked with the Raiffeisen Group again and again in 2020 and 2022. In the first phases, we developed and implemented the use case for their first proof of concept for a chatbot to answer customer inquiries about Twint. My role was to support the team in finding and designing the use case, creating the dialogs and conducting target group workshops to find the right chatbot tonality.

In the subsequent project, I worked with my team to develop and define all the technical and content requirements for the planned conversational banking project. Specifications and requirements were drawn up and all the necessary documents for a proper tender were compiled. My team then carried out the entire tender process up to the selection of the right provider. This also includes the pre-selection of which providers are invited, as well as the preparation of the pitch dates and the creation of the selection criteria.

Challenges

• Very high data protection requirements
• Many and complex internal systems for which interfaces are needed

Highlights

• Very extensive and detailed tender documents
• A well thought-out selection process with a precise list of criteria

LLM chatbots for the relief organization of the Protestant Church (HEKS)

The aid organization of the Protestant Church (HEKS) receives many requests, both internally and externally. As part of a general training course, I trained the entire team on generative AI and chatbots. We then defined three use cases for LLM chatbots. These mainly include internal applications. I then selected the right technology provider to implement these applications and supported the implementation. In addition to technical issues, my task also included the usability of the bot. I tested the chatbot and gave feedback on optimizations.

Challenges

• Very broad and diverse team at HEKS
• Many different use cases for LLM chatbots
• Limited budget

Highlights

• Training at various levels of knowledge
• Selection of a suitable technical solution that not only fulfills all technical requirements, but also fits the HEKS budget.

Migration of the chatbot of the Sparkassen-Finanzgruppe

2021 – 2022 I worked very closely with the chatbot team of the Sparkassen Finanzgruppe in Germany. My task was to migrate the existing chatbot to a new technology. First of all, a proper concept had to be created so that nothing could go wrong during the migration of over 1400 intents. I then carried out the majority of the migration myself. The migration was carried out under the motto “Voice First”, which means that all chat dialogs also had to be prepared for a voicebot.

Challenges

• A variety of intents and exceptions
• High data protection requirements
• Interest in voice-first dialogs

Highlights

• Creation and implementation of a successful migration plan
• Redesign of all dialogs in Voicefirst
• Reorganization of many dialogs

Migration and optimization of the rule-based chatbot to the LLM chatbot of Luzerner Kantonalbank

Luzerner Kantonalbank initially introduced a rule-based chatbot independently with the help of a SaaS solution. However, this quickly reached its limits and an LLM chatbot was used to increase the knowledge of the bots. My task was to provide the team with the necessary knowledge about LLM chatbots and then support them in developing the migration roadmap. During the migration, I repeatedly carried out usability tests and gave the Luzerner Kantonalbank team and the technology provider tips on how to improve the LLM chatbot.

Challenges

• High data protection requirements
• Migration of a rule-based chatbot to the LLM chatbot

Highlights

• Successful migration of the first use cases
• Significant optimizations with regard to the usability of the LLM chatbot
• Precise and well thought-out prompt engineering

An LLM chatbot for the coffee machine manufacturer Jura

The coffee machine manufacturer receives a large number of support and advice requests. This often involves very specific questions for which employees need to know the exact specifications and instructions for the individual coffee machines. These are sometimes challenging questions, even for an LLM chatbot. There must be no misunderstandings. My main task in this project is to define all the technical and content requirements for a functioning LLM chatbot and to find the right technology provider. In addition to answering the questions correctly, the flexibility for further interfaces to internal systems is also an important challenge.

Challenges

• The Jura team had no previous experience with chatbots
• Many special cases in terms of content
• Many technical terms that can lead to misunderstandings for the LLM

Highlights

• Sophisticated preparation of content for the LLM
• Precise selection of the right technology provider

Chatbot concept workshop and implementation with the NGO “Pusch – Practical Environmental Protection”

Pusch is a Swiss NGO for environmental protection. In a very detailed chatbot workshop, we defined and prioritized various possible uses of chatbots within their customer journey. For the implementation, I supported the team in choosing the right technology and suitable software partners.

A chatbot for the HR department at Liebherr

Liebherr is one of the largest manufacturers of construction machinery in the world. I accompanied the HR team during the introduction of their first chatbot. The main aim was to define all the requirements and shortlist suitable providers and ultimately find a provider for the B2B company. Data protection was usually the top priority, followed by criteria for internal interfaces.

The chatbot for Lucerne Cantonal Hospital

Lucerne Cantonal Hospital started to gain initial experience with chatbots very early on. In 2020, I implemented the Swiss hospital’s first chatbot. My first task was to narrow down the exact use case and develop the entire concept for the chatbot. I then defined all the technical and content requirements and found the right technology partner as part of a small tender process. With the help of these partners, I then implemented the entire chatbot and also the introduction, including the Communication measures accompanied

An AI chatbot for the employees of Schulthess Maschinen AG

Schulthess Maschinen AG is one of the largest suppliers in the field of washing technology. Many customer inquiries are repetitive and are therefore ideally suited for the use of chatbots. As Schulthess Maschinen AG itself has no expertise in chatbots, I initially provided the team with general training on chatbots. We then worked together to define and prioritize the most important use cases for an LLM chatbot. In addition to use cases for end customers, the main focus was on use cases for internal chatbots. Once the right technology had been selected, the first chatbots were introduced for internal use.

The chat personas of Baloise Bank SoBa

Baloise Bank SoBa has been experimenting with chatbots in e-banking since 2020. But how should the bots communicate? What tonality are they allowed to use? We worked on these and other questions in workshops lasting several days with the chatbot team at Baloise Bank SoBa and developed a “How to Bot Guide”. This guide was developed specifically for the banking team and is constantly being developed further.

With the right chatbot for further education at the University of Applied Sciences of the Grisons (FHGR)

The Graubünden University of Applied Sciences wanted to use chatbots to raise awareness of individual degree programs. The digital assistant should also be able to answer students’ most important questions around the clock. I managed the entire chatbot project and also implemented the bot with the help of a suitable SaaS solution. In addition to selecting the right chatbot software provider, the main focus here is on dialog design. The dialogs must motivate users to register for a degree program at the university of applied sciences. This often involves psychological components and not just technical requirements.

Even more references

Of course, it’s best if you write to me personally and tell me what kind of references you’re looking for. I will be happy to send you further information.

You are very welcome to contact me personally. Just send me a message – preferably by WhatsApp message or e-mail.