Ultra-High-Performance Smart Home Assistant for Raspberry Pi 5

A next-generation smart home assistant optimized for Raspberry Pi 5, leveraging local language models via Ollama for robust natural language understanding and device control. Designed for low latency and high throughput.

🚀 Features

• LLM-First Architecture: All commands are processed by a local language model, enabling flexible and conversational interactions beyond traditional rule-based systems.
• High Performance: Engineered for Raspberry Pi 5 with memory-mapped caching and multi-threading, delivering 30+ tokens/second. Lightweight and responsive, even on resource-constrained hardware.
• Model Flexibility: Supports multiple optimized models, including tinyllama:1.1b, qwen:0.5b, and deepseek-coder:1.3b. Includes a benchmarking tool to compare model performance.
• Real-Time Control: Direct GPIO management for smart devices using gpiozero and the modern lgpio pin factory, ensuring low-latency hardware communication.
• Web Dashboard: Responsive web interface
• System Optimizations: Options to apply system-level enhancements for Raspberry Pi, such as setting the CPU governor to "performance" and increasing process priority.

🏛️ System Architecture

This assistant uses a modular, LLM-first architecture for maximum performance and flexibility.

• Input Handling: Accepts user commands via interactive terminal or FastAPI web interface.
• Command Understanding:
  • NEON Path (Fast Path): Simple commands (e.g., "turn on the living room light") are processed using a NEON-optimized, hardware-accelerated string matcher for instant recognition and execution.
  • LLM Path (Flexible Path): Complex or conversational commands are sent to a local Ollama instance. The language model interprets intent and returns a standardized JSON object for action.
• Action Execution: Parses JSON output to determine and execute device actions or scene activations.
• Device Control: Interfaces directly with Raspberry Pi GPIO pins via gpiozero for smart device management.
• State Management: Tracks device and sensor states in real time for an accurate smart home overview.
• Web Dashboard: FastAPI server provides a real-time dashboard accessible from any device on your network.

✅ Requirements

Install the following Python libraries:

ollama
numpy
httpx
orjson
gpiozero
fastapi
uvicorn
psutil
uvloop

A running Ollama instance with your chosen models is required.

🛠️ Installation

Clone the repository:

git clone https://github.com/fidel-makatia/EdgeAI_Raspi5.git
cd EdgeAI_Raspi5

Create and activate a virtual environment:

python3 -m venv venv
source venv/bin/activate

Install dependencies:

pip install -r requirements.txt

Install Ollama:
Follow the official instructions for Raspberry Pi: Ollama Installation Guide

Download a language model (example):

ollama pull tinyllama

⚙️ Usage

Run the main script:

python3 smart_home_assistant.py --model [MODEL_NAME]

Command-Line Arguments:

• --model (Required): Name of the Ollama model (e.g., tinyllama, qwen:0.5b)
• --port: Web server port (default: 8000)
• --no-web: Disable web interface (terminal-only mode)
• --optimize: Apply system-level optimizations (requires sudo)

Example:

python3 smart_home_assistant.py --model tinyllama --optimize

Access the web dashboard at:
http://<your-pi-ip>:8000

📝 Example Commands

Here are some example commands you can use with the assistant:

• turn on living room light
• I want to watch my favorite show
• its getting late, secure the house

📊 Performance Benchmarks

Model	Tokens/Sec	Avg Latency	Time/Cmd
qwen:0.5b	17.0	8217	8217
tinyllama:1.1b	12.3	9429	9429
deepseek-coder:1.3b	7.3	22503	22503
gemma2:2b	4.1	23758	23758
deepseek-r1:7b	1.6	64797	64797

Looking through the code, I can extract the GPIO pin assignments from the _init_devices() method. Here are the Raspberry Pi pins used:

🔌 Raspberry Pi GPIO Pin Assignments

Device	GPIO Pin	Type	Room	Dimmable
Living Room Light	GPIO 17	Light	Living Room	✅ Yes
Living Room Fan	GPIO 27	Fan	Living Room	❌ No
Smart TV	GPIO 22	Smart TV	Living Room	❌ No
Bedroom Light	GPIO 23	Light	Bedroom	✅ Yes
Bedroom AC	GPIO 24	AC	Bedroom	❌ No
Kitchen Light	GPIO 5	Light	Kitchen	❌ No
Front Door Lock	GPIO 26	Door Lock	Entrance	❌ No
Garden Light	GPIO 16	Light	Outdoor	❌ No

📋 Pin Summary

• Total Pins Used: 8
• Dimmable Devices: 2 (use PWMLED)
• Standard Devices: 6 (use LED)

🔧 Hardware Requirements

• Raspberry Pi 5
• 8x Relay modules (or similar switching devices)
• 2x PWM-capable outputs for dimmable lights
• Proper power supplies for high-wattage devices (especially the 1200W AC unit)

⚠️ Safety Notes

• Use appropriate relays rated for the device power consumption
• Always use proper isolation between Raspberry Pi GPIO (3.3V) and mains voltage devices
• Consider using optocouplers for additional safety

🤝 Contributing

Contributions are welcome! Please submit a pull request or open an issue for improvements or bug reports.

📄 License

This project is licensed under the MIT License. See the LICENSE file for details.