Model-based systems engineering (MBSE) is increasingly gaining momentum in industries thatare developing and managing complex systems, such as Volvo Construction Equipment (VCE).Transitioning from informal architecture documentation, such as ad-hoc sketches or diagrams,remains a challenge, especially due to steep learning curve and complexity of modeling languages such as SysMLv2. This study explores how a natural language template can help domain expertsintuitively capture system architecture information while enabling transformation of these templates into formal SysMLv2.To achieve this, two artefacts were introduced: a natural language template and an LLM driven transformation pipeline. The natural language template was iteratively evaluated and refined through a collaboration with a company representative, ensuring that it is both intuitive and sufficiently structured. The transformation pipeline uses the provided template as input and generates SysMLv2 code that is then rendered into a graphical representation of the system. The results show that the template successfully captures system architecture information while maintaining informality and intuitiveness. The transformation pipeline consistently generates SysMLv2 models, albeit with a missing variability concept due to the current limitation of the LLM training on SysMLv2 syntax. This work demonstrates the feasibility of bridging the gap between informal and formal modeling, lowering the entry barrier to SysMLv2 and enforcing MBSE practices by enabling a flexible modeling approach.