Artificial olfactory systems have been applied in domains such as food quality assessment, environmental monitoring, and medical diagnostics. However, progress in enabling machines to perform high-level reasoning based on odor perception remains limited. To address this gap, we propose a novel hybrid system that integrates electronic nose (E-Nose) sensing with large language models (LLMs) and knowledge graphs, enabling human-like olfactory reasoning through the interaction of sensory and linguistic data. A case study on coffee aroma interpretation demonstrates the system’s ability to generate descriptive narratives, infer semantic relationships, and contextualize odor signals meaningfully. To simulate odor perception, we employed a TETCN model—combining a transformer encoder and a temporal convolutional network—to predict aroma types and generate structured labels. These labels guide the retrieval of relevant knowledge from a memory database, which is then processed by the LLM for advanced reasoning. By bridging signal-level perception and abstract cognition, this work presents a significant advancement toward cognitively intelligent olfactory systems.