This thesis investigates the use of large language models (LLMs) for analysing Internet of Things (IoT) data. IoT systems collect large volumes of heterogeneous sensor data that are hard to interpret without technical expertise. We systematically evaluate six large language models on the task of translating natural-language questions into SQL queries over three public IoT datasets, using 52 standardised queries and five metrics (accuracy, SQL validity, error rate, latency and cost), complemented by a separate visualisation benchmark. Building on these findings, we develop the ChatDB conversational prototype, which combines natural-language-to-SQL translation, automatic visualisation and result explanation in an interactive interface.
|
