A chatbot that explains complex railing guidelines in an understandable way? This is not only possible, but has already been successfully realised. In collaboration with the Bundesverband Metall, DAISEC has developed a specialised chatbot that provides precise answers to questions about the extensive railing guidelines.
Challenges with complex documents
The railing guideline of the Bundesverband Metall is a technical document that defines the standards and requirements for the construction, installation and safety of railings in various scenarios. It includes detailed instructions and technical specifications to ensure that all railings meet the highest safety and quality standards.
The more than 50-page guideline with its numerous tables and cross-references placed special demands on the technical implementation of the chatbot. The usual OpenAI Assistants API, which is well suited for simple customer enquiries, reached its limits here. The DAISEC experts therefore opted for an advanced approach with established RAG frameworks (Retrieval Augmented Generation), which enable more precise processing of extensive documents.
Innovative technical solution through collaboration
The system was evaluated in close cooperation with a terrain expert from the federal association. To this end, the team developed specific questions to continuously analyse and improve the quality of the answers generated.
Technical development and optimisation
The response quality has been continuously improved through various measures:
- Systematic evaluation of the context to identify sources of error, thereby optimising the relevance of the document sections provided;
- Continuous adaptation of the machine-readable version of the policy to ensure optimal processing by the AI system;
- Integration of modern open source language models such as Llama 3.3 and Deepseek-V3, which enable larger context windows at lower costs.
User-friendly implementation
Following successful optimisation, an intuitive web interface was developed for using the chatbot. Users can use a sidebar to select important parameters such as federal state, fall height and type of housing construction, as these influence the requirements for the railings needed and are changed more frequently. The chatbot thus provides customised answers to the required guidelines based on the parameters. An integrated feedback system enables continuous monitoring of the answers with the help of user feedback.
Successful validation of the concept
The project has proven to be a successful proof of concept for a policy chatbot that has convinced even sceptical experts of the quality of the answers. Even though features such as complex database schemas, rate limiting or automated A/B tests were not used, the project clearly demonstrates the potential of AI-supported assistance systems for the interpretation of complex guidelines.
Possible applications
The chatbot developed demonstrates how artificial intelligence can make complex technical guidelines accessible. This opens up prospects for similar applications in other technical areas where precise interpretation of regulations is required.
Your DAISEC contact
Falls Sie Fragen haben oder mehr darüber erfahren möchten, wie Sie KI-Chatbots in Ihrem Unternehmen einsetzen können, stehen wir Ihnen gerne zur Verfügung, um Ihnen den Einstieg in diese zukunftsweisende Technik zu erleichtern.
Expert for Artificial Intelligence
