Аннотация
This work introduces MicroTrack, an intelligent interpretation system for precision hydroponics that addresses the semantic gap between raw Internet of Things (IoT) telemetry and the decisions that non-expert operators are required to make. Conventional precision agriculture relies heavily on graphical dashboards, which present every relevant variable but implicitly assume a level of domain expertise that the target users do not possess. We present a hybrid interface that augments a multi-modal 12-channel sensor network with a localized Large Language Model (LLM) acting as a reasoning engine. MicroTrack employs dynamic context injection within a Retrieval-Augmented Generation (RAG) framework to ground the model in the live state of the grow-box and to translate numerical time-series into articulated, actionable recommendations. The system was evaluated through a within-subjects pilot user study with four non-expert participants and sixteen trials, using Latin-square counterbalancing across four scenarios that included two adversarial cases, namely a benign all-normal state and a physically impossible sensor reading. In this pilot, the Hybrid interface reduced mean decision time from 221.16 s to 15.52 s, corresponding to a Cognitive Load Reduction Factor of approximately 14.25, and raised accuracy from five of eight to eight of eight correct decisions. It also prevented two non-expert failure modes that the visual-only baseline failed to suppress, namely cognitive over-attribution on benign states and unconditional trust in malfunctioning sensors. These findings suggest that grounded linguistic interpretation, layered on top of existing IoT dashboards, can be an effective mechanism for widening access to complex agricultural infrastructure for non-expert urban cultivators.