Scholarlink

This paper presents a comprehensive study on integrating large language models (LLMs) with modern smart home automation systems. Building upon the foundation provided by the home-llm project, our approach leverages Home Assistant as the central automation platform while interfacing directly with state of the art LLMs, Polka 1.1B and Bielik 11Bv2. Additionally, we explore the potential of using Boneio controllers based on the ESP32 microcontroller as a dedicated execution layer, bridging the gap between digital command processing and physical device control. Furthermore, we introduce a finetuning methodology using a custom dataset generated via a proprietary prompt generator, enabling the creation of over 10,000 sample prompts. Polka 1.1B based model is finetuned using a standard process, while Bielik 11Bv2 is adapted using the Low-Rank Adaptation (LoRA) technique. This work outlines a unique, scalable architecture that enhances natural language command interpretation and achieves efficient realtime execution in smart home environments.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Integration of Polish Large Language Models for Smart Home Automation

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Michał Pikus,
  • Jarosław Wąs,
  • Agata Kozina

摘要

This paper presents a comprehensive study on integrating large language models (LLMs) with modern smart home automation systems. Building upon the foundation provided by the home-llm project, our approach leverages Home Assistant as the central automation platform while interfacing directly with state of the art LLMs, Polka 1.1B and Bielik 11Bv2. Additionally, we explore the potential of using Boneio controllers based on the ESP32 microcontroller as a dedicated execution layer, bridging the gap between digital command processing and physical device control. Furthermore, we introduce a finetuning methodology using a custom dataset generated via a proprietary prompt generator, enabling the creation of over 10,000 sample prompts. Polka 1.1B based model is finetuned using a standard process, while Bielik 11Bv2 is adapted using the Low-Rank Adaptation (LoRA) technique. This work outlines a unique, scalable architecture that enhances natural language command interpretation and achieves efficient realtime execution in smart home environments.