Engineering design processes for complex products utilize and generate large amounts of heterogeneous data, ranging from structured product, part and supplier specifications all the way to unstructured sources such as reports, test and maintenance logs, change orders as well as user feedback. These data sources often remain siloed, limiting opportunities for integrated analysis and knowledge re-use. This work explores the potential of an end-to-end Multi-Agent System (MAS) based on Large Language Models (LLMs) for analyzing and selecting engineering data, generating Knowledge Graph (KG) schemas and performing entity resolution for graph linking to address this challenge. We introduce an overall workflow based on a unified three-layer graph architecture consisting of a reference KG (structured engineering data), an observation KG (entities and relations extracted from unstructured sources), and an evidence KG (text chunks providing provenance) and demonstrate how specialized agents can collaborate to construct it. A proof-of-concept case study using a student drone design course illustrates how design data and user feedback can be processed into the three-layer unified graph structure serving as a substrate for using GraphRAG to illustrate the possibility of answering complex engineering questions. The results demonstrate the promise of LLM-based agentic systems to increase automation and flexibility in dynamically gathering and unifying data across formats for KG construction, establishing a foundation for downstream applications grounded in engineering data enabling explainable, AI-driven reasoning in product development contexts. Finally, current limitations and further research directions are discussed.