We interviewed our colleague Geoffrey Decan about the AI chatbot, called “Lexi”, which is available from Diamonds 2026 onwards. With the launch of Lexi, BuildSoft is taking an important step in making AI accessible within Diamonds. In this interview, Geoffrey explains why we deliberately chose to develop our own solution, how Lexi works behind the scenes, and what our vision is for the future of AI within Diamonds.
Geoffrey has been working at BuildSoft since 2011. Since April 2025, he has been part of the AI initiative within the StruSoft Group as BuildSoft’s representative. Since January 2026, he is AI Lead and Technical Analyst for the StruSoft Group. He works from our Spanish office in Alicante. Read on to discover what he has to say on this topic departing from his extensive experience.
1. Why did we choose a chatbot?
When using Diamonds, you often run into questions such as: “Where was that setting again?”, “Can I perform this calculation?”, “Which workflow do you recommend?”, and so on.
In the past, this usually meant searching through the manual or opening a browser and looking it up on the support website. But that takes time and breaks your concentration.
With Lexi, you start that search directly in Diamonds: you ask your question in the sidebar and immediately receive a practical answer — including a reference to the relevant documentation so you can read further if needed. Lexi also shows formulas and images so you can instantly verify the answer.
In short: Lexi reduces context switching. You move forward more quickly while still being able to rely with confidence on the original source information.
2. For which type of engineer is Lexi most valuable: junior, senior, or both?
Lexi is for everyone. For junior users, Lexi shortens the learning curve and increases independence. It helps explain parameters, interpret error messages, and provides step-by-step guidance when setting up an analysis.
For senior users, Lexi saves time and deepens expertise. It can confirm exactly what the software does according to the documentation. Lexi lowers the threshold for working with advanced functionality.
Lexi is not a replacement for expertise, but an accelerator of it — whether you are just starting out or have been using our software for years.
3. Lexi is a newly developed chatbot. Why did we deliberately choose to develop something new rather than use an existing chatbot such as ChatGPT?
We wanted a fully integrated solution within Diamonds that focuses exclusively on content-related questions about Diamonds, rather than a general-all-purpose tool. This allows us to provide targeted, product-specific support within the user’s familiar working environment.
By opting for in-house development, we also retain control over Lexi’s content. While general chatbots rely on a mix of sources that may be outdated, incomplete, or even related to other software packages, Lexi deliberately runs on our own knowledge base.
And perhaps the most important reason: privacy. It is not always clear what a general chatbot does with user data. By developing Lexi entirely within our own environment, we ensure that customer data remains protected and is not shared with external parties.
4. Can you clarify Lexi’s privacy settings?
Lexi runs in a data centre in Sweden. However, conversation content is stored locally on your device — not in Sweden and not in the cloud. You decide which data you share with us to improve Lexi. The table below compares the three privacy settings. For clarity: data is only shared with us when you provide feedback.
| Which data? | Privacy settings of Lexi | ||
|---|---|---|---|
| Nothing | Minimal | Full | |
| Feedback (👍 or 👎) | Not possible to give feedback | Yes | Yes |
| Comment | Not applicable, as data is only sent when you give feedback | Not possible to add comments | Yes |
| Title of the conversation with Lexi | Yes | Yes | |
| Content of the conversation with Lexi | No | Yes | |
| Personal data (name, email, phone number, licence number, etc.) included in your conversation with Lexi | Only anonymised feedback is shared with us. We have no way of identifying your personal data and therefore cannot contact you. If you wish to contact support, please email or call us. | ||
| Your model (*.bsf) | Lexi does not access your model or its results. It will therefore never share your model with us. If you are unable to resolve a model-related question, please email your model (*.bsf) to support@buildsoft.eu | ||
5. What happens behind the scenes in Lexi?
You can think of Lexi as two components working together: a language model and a smart search layer.
The language model is essentially a highly trained text prediction system. Think of your phone trying to predict the next word as you type a message — but far more advanced.
The language model has learned how sentences are structured, which words logically follow one another, what explanations, summaries and dialogues look like, and how different writing styles are constructed. When you ask a question, the language model analyses the patterns in your query. It predicts the answer word by word. This happens extremely quickly, resulting in a fluent response.
However, a language model does not “know” things in the way a human does. It continuously calculates the most probable next word. Yet because it has seen so many examples, it feels as though it understands what you are saying.
The language model does not have live access to up-to-date knowledge or information. You need a smart librarian — the search layer — which retrieves relevant information from all available sources and passes it on to the language model. The language model can then formulate a clear answer to your question.
Specifically for Lexi, the workflow is as follows:
- You ask Lexi a question.
- Lexi reformulates, adjusts or summarises your question within the context of Diamonds and/or previous conversations or data. This helps the AI better understand what is relevant.
- Lexi performs a semantic search (based on meaning, not just keywords) in the available sources. If you use the word “beam” in your question, Lexi understands that you might also mean “girder”.
- Lexi has access to the entire support website (= a knowledge base). Since 2025, the support website also includes the full Diamonds reference manual. From all articles related to the requested topic, Lexi selects the best matches.
- Lexi formulates an answer and returns it to you together with source references.
6. What type of questions can you ask Lexi?
Lexi is at its strongest when it comes to product-related questions about Diamonds: settings, workflows, where to find something, step-by-step instructions, explanations of features, and so on.
Lexi is not a replacement for engineering judgement. You cannot ask it to interpret a standard (e.g. Eurocode) or debate whether element X complies or not. Lexi is also not capable of looking “inside” your model or results.
7. How do we avoid “hallucinations”?
Hallucinations are confident but incorrect answers — a known risk with language models.
Lexi is deliberately scoped: it is knowledge-base-driven and never searches for information outside that knowledge base. In addition, we apply quality controls and automated evaluations to test and improve answers.
If Lexi cannot find sufficiently strong sources, it will indicate that no suitable answer could be found. This may mean the information does not yet exist or that your question needs to be phrased differently.
Practical tip: try rewording your question, using a different formulation, or adding a key words (for example the name of a module or dialogue window). And if the topic is genuinely missing, that is valuable feedback for us to further expand the documentation.
8. Is Lexi the final destination of our AI journey?
Absolutely not. Lexi is our starting point for making AI available in a practical way within Diamonds: today, we have a chatbot that can answer product-related questions. In the short term, we will refine this further — better answers, more and up-to-date knowledge base content. But the ultimate goal is to expand Lexi so it can handle more complex scenarios:
- Phase 1 (now): product knowledge
Lexi answers product-related questions about Diamonds, with source references. - Phase 2 (planned 2027): model knowledge
Lexi can look “inside” your model and analyse its contents, answering questions such as: “Which profiles are used in this model?” - Phase 3: smarter reporting workflow
Lexi can execute instructions related to reporting features. For example: “Create a report based on this template, and also include an image of the vertical deformations in SLS RC.” - Phase 4: simple commands and automation of multi-step calculations
Lexi can execute commands such as: “Change the concrete grade from C25/30 to C30/37, rerun the elastic analysis and calculate the reinforcement.”
Our ambition is clear: to deploy AI where it truly adds value — reducing complexity, removing friction, and helping engineers reach better decisions more quickly. Lexi is therefore not a final destination, but a foundation on which we will continue to build in the years ahead.
