Dlubal uses CustomGPT.ai

Transforming Customer Support: Dlubal’s AI-Powered Engineering Solution

Use Case

Proof-of-Concept

Industry

AI Development

Features

Data Integration, Anti-Hallucination, Web Deployment, Security, Sitemap Integration, No-Code Platform, Custom Persona

At Dlubal, we are constantly looking for ways to optimize our processes and improve the customer experience. The development of an assistant using Customgpt's facilities was a crucial step forward. The assistant has enabled us to offer 24/7 support while improving accuracy and speed of response. This has led to a noticeable increase in customer satisfaction and even faster support. At the same time, our support team has seen a significant increase in the efficiency of our customer service. The assistant provides precise answers in real time and significantly reduces the workload of our support team.

George Dlubal, CEO at Dlubal Software

Goals

  • Website-based 24/7 customer support chatbot and civil engineering assistant
  • Utilize a large language model (LLM) but ground the bot with company data
  • Scalability with excellent API capabilities to integrate with company-built software used by more than 13,000 companies

Approach

  • Index JSON and PDF files and an extensive sitemap including e-learning materials
  • Expertly fine-tune the assistant Persona, expertise, and language delivery
  • Embed AI assistant “Mia” to the company website and in the software the company provides to customers
  • Continuously fine-tune based on insights from the likes/dislike feature, chatlogs and customer feedback

Result

  • Mia is now providing customer and technical support to over 130,000 software users
  • Dlubal hopes to extend Mia’s already impressive capabilities and explore further use cases, including marketing

About Dlubal

Dlubal Software is a global leader in advanced structural analysis and design software used by structural and civil engineers. The family-founded and run company is headquartered in southern Germany and has sales, marketing, and technical development offices worldwide, including a substantial technical presence in Prague, Czech Republic, and offices in the U.S. Dlubal’s core programs are used in 132 countries. With 35 years of experience, Dlubal serves 13,000 companies and over 130,000 users with its high-precision products.

Prof. Dr. Michael Kraus is an expert in machine and deep learning. He works closely with Dlubal founder, CEO, and technical lead Georg Dlubal, and together with Dogukan Karatas and the AI team they have been developing Dlubal’s AI strategy and products.

Dlubal's Goals

Dlubal has several technical engineers providing support to customers. Inquiries can range from software issues and handling bug reports to sales and invoicing questions. Like many technology companies, Dlubal struggles with the global talent shortage and is constantly hiring. 

The need to provide 24/7 quality customer support and fill gaps in its skilled workforce led Dlubal to begin investigating the use of generative AI and, specifically, large language models (LLMs) for customer support. However, this isn’t Dlubal’s first foray into AI. The company has been developing its own machine-learning deep architectures for the past several years. This experience gives the team an inherent understanding of the advantages of generative AI and pre-trained models. 

Beginning in 2023, Dlubal saw the project as a test stage to discover the quality of the content that could be generated from the information Dlubal had internally, as well as what’s publicly available. Dlubal's AI team created a prototype, heading in the direction of retrieval augmented generation (RAG) technology but ran into issues of scalability. This led Dlubal's AI team to research vendors that would allow them to use an LLM but ground it in their own data and integrate it not only into the company website but also into Dlubal software.

We looked at different vendors and in the end, we chose CustomGPT.ai because for us, it had the best spectrum of quality, of answers, ease of use, scalability, and most importantly, we also need a lot the API capabilities. We have some internal processes so we want to rely on an automated connection to the chatbot, to the LLM system and the (CustomGPT.ai) API offers great value.

Michael Kraus

AI Expert at Dlubal

CustomGPT.ai’s API capabilities allow Dlubal to automate certain processes and deeply embed the bot into its systems and software to achieve significant internal value. 

Dlubal further envisioned the chatbot becoming a general expert in civil engineering. Visitors could ask complex questions, and the bot would even provide the standard formulas for structural calculations, creating an engaging tool on the website. The software company was keen to demonstrate to its customers that, as a brand, it had AI on its mind and an AI agenda.

Dlubal's Approach

Dlubal's AI team created Dlubal’s CustomGPT.ai agent, or AI assistant Mia, and implemented it quickly into dlubal.com. They first prepared and compiled company data into specific files in JSON and PDF format before uploading via CustomGPT.ai’s interface. Much of Dlubal’s information, background documentation, tutorials, and extensive e-learning guides can be found on the website, so using the website content by simply adding the sitemap was a vital part of “training” Mia. 

The next step was to fine-tune Mia’s persona. It was critical to founder Georg Dlubal that Mia’s language and style met their requirements and that on the technical side, for example, formulas were rendered correctly. 

The team spent an intense two weeks working closely with CustomGPT.ai. They wanted to avoid certain behaviors from the assistant but foster others and to make the bot available on the website and in its integration with Dlubal software in ten languages but with accurate autography and accents. Mia also had to learn about Dlubal software and its product versioning.

We also worked for a week or more on integrating the chatbot into our own software. We had some issues with the sizing and wanted to use the rest method to override, for example, the snippet you provided so that we could use multiple languages. So for us. We saw some of the issues we were always in an open dialogue, especially with Eli. What really is a great plus, and it’s also the opinion of our (CEO) is that you provided constantly really quick and very nice support.

Michael Kraus

AI Expert at Dlubal

After extensive testing and refining of Mia’s Persona, knowledge, and capabilities, the Dlubal team now spends a few hours a week continuing to fine-tune this new and important business tool. 

Dlubal found the ability of Mia’s users to like or dislike responses particularly insightful for evaluating the AI’s performance and developing a deeper understanding of certain customers. The ratio of dislikes compared to the overall number of queries was a first indicator of what needed to be improved, perhaps in how Mia was fed data or if a specific piece of software knowledge was missing. 

Critically, Dlubal found a reason for the dislikes was that the answer Mia gave was technically correct but the customer was expecting a different outcome. Dlubal has different product versions, and some older products are no longer supported. The company policy is to push its new products as norms and designs in the industry are changing, and Dlubal today has a different philosophy on how to use its products. For example, Dlubal provides API access to customers so they further interact and build the capabilities of Dlubal software and their own systems.

Some of Dlubal’s customers in the field prefer the older software versions, so the dislikes given to Mia’s responses when, in fact,  technically correct became a great indication of this issue and the opportunity for Dlubal to contact these customers in person to convince them of the benefits of upgrading to the company’s new products. 

Dlubal's Results

Mia, Dlubal’s incredibly well-trained and knowledgeable AI assistant, is completely operational at dlubal.com, answering technical software and administrative customer questions and providing an impressive civil and structural engineering resource (with citations) for Dlubal’s 130,000+ users and website visitors. Mia has also been embedded into Dlubal software to serve customer needs right where they require the most support. 

Dlubal’s next use case for CustomGPT.ai and Mia will be in marketing. However, the close partnership formed between Dlubal and CustomGPT.ai sees the Dlubal team hoping to see even further generative AI functionality from the platform in the coming months, ideally in the form of voice or image developments so, for example, Mia could support structural image rendering in the future, with the right fine-tuning.

To get in touch with Dlubal, please contact Daniel Dlubal at daniel.dlubal@dlubal.com.

To read other Case Studies like this one, visit https://customgpt.ai/customers/.

Want to learn more about CustomGPT?

Sign up here or get in touch with one of our Customer Success Agents to get started.