Small Language Model-Based Smart Home Devices for Human-Building Interaction

Project Overview

This project explores the integration of small language models (SLMs) with smart home environments by deploying the Phi-3 Mini model and Llama 3.2 onto a Raspberry Pi 5 to enable natural language interaction with building systems. The primary objective is to enhance human-building interaction through accessible, cost-effective, and localized AI-powered solutions. This project seeks to revolutionize human-building interaction by embedding generative AI within the built environment. By enabling localized AI-driven decision-making, the research contributes to the development of more responsive, secure, and sustainable smart home ecosystems, forming the backbone for future smart cities.

Summary

Video Demo

Click on the image to view the video.

Goals and Objectives

• Human Building Interaction: Facilitate intuitive voice-controlled management of smart building systems, such as lighting, temperature regulation, and other IoT-connected devices.
• Edge AI Deployment: Leverage low-cost single-board computers like the Raspberry Pi 5 to run open-source SLMs locally, eliminating dependency on commercial cloud-based AI services.
• AI-Driven Decision-Making: Integrate advanced AI capabilities directly into building infrastructure to support efficient, real-time decision-making.
• Cost-Effectiveness: Lower the barrier to entry for AI adoption in smart homes by using open-source tools and affordable hardware.
• Scalability: Establish a foundation for scaling edge AI solutions from smart homes to broader smart city initiatives.

Broader Impacts

• Smart Building Evolution: Accelerates the transition toward AI-empowered building systems, fostering smarter, more sustainable infrastructure.
• Smart City Integration: Demonstrates a scalable model for applying generative AI to interconnected urban environments, promoting responsive and efficient smart city initiatives.
• Democratizing AI: Makes advanced AI solutions accessible to a broader audience by utilizing open-source models and affordable hardware.

Manuscript

Under preparation...

Requirements

• Open-source Large language model (e.g., LLaMA)
• Generative AI inference tool. llama.cpp
• Python 3.10
• Raspberry Pi and IoT sensors
• Open-source Text-to-Speech (TTS) model, Whisper
• Open-source Speech-to-Text (STT) model, Piper
• Smart home devices (e.g., smart lights, smart fan, smart humidifier, smart air purifier)

Detailed setup guide

Coming soon.....

License

This project is licensed under the MIT License.