Technology-driven companies often express a strategic ambition toward customer-centricity, but struggle to translate it into operational practices such as solution discovery. In business-to-business-to-consumer (B2B2C) ecosystems, where organizations reach end-customers indirectly through business partners, commonly used design tools such as the Strategyzer Value Proposition Canvas offer limited guidance. While solutions in these contexts should be shaped by customer insights, internal limitations and partner-imposed pressures often override them. This paper presents a practice-based case study of how the Value Proposition Canvas was extended within a cybersecurity organization undergoing a customer-centric transformation. Following a research-through-design approach, the canvas was iteratively refined through co-creation workshops and reflection. The final version incorporates internal and partner pressures, desired customer behaviours, explicit opportunity framing, a user story element, and early acceptance criteria. The findings show how the extended canvas helps teams move from strategic intent to concrete, customer-informed solution definition in complex B2B2C environments.

In Requirements Engineering, information is often captured in unstructured natural language, limiting its systematic use across product development phases. Ontologies offer a machine-interpretable framework that integrates domain-specific knowledge. This study presents a text-to-ontology pipeline combining a rule-based NLP approach using BERT and a generative LLM (GPT), demonstrated on an electrified micro-mobility solution. Both enable semi-automated semantic integration, with GPT offering stronger contextual abstraction and BERT providing stable results for syntactically simple input.

3D Computer-Aided Design (CAD) reverse engineering aims to recover the parametric command sequences required to reconstruct 3D geometries from point clouds, exchange formats like STEP, or triangular meshes. Current Large Language Model (LLM)-based methods that generate CAD code rely on single-shot test-time sampling, often failing on complex topologies. This paper proposes Tree-of-CAD Models (ToCAD) to empower reverse engineering LLMs with frontier multimodal LLMs for iterative visual refinement in a search-based framework. By intelligently guiding test-time compute, ToCAD significantly improves the topology of reconstructed 3D models, improves code structure, and enables the utilization of complex CAD features beyond sketch and extrude operations.