Raspberry Pi

Train a custom Thai wake word with Porcupine and parse Thai voice commands locally with Rasa NLU to control smart home devices — all audio processing stays on-premises with zero cloud dependency.

Build an acoustic anomaly detection system using YAMNet TensorFlow Lite on Raspberry Pi 5 to detect glass breaking, screaming, and alarm sounds, triggering Home Assistant alerts and LINE notifications.

Design a unified smart home dashboard architecture using Mosquitto MQTT broker, Node-RED for data flow, InfluxDB time-series database, and Grafana for visualisation — consolidating security, air quality, and elderly monitoring data into a single dashboard.

Use an LSTM neural network to forecast Bangkok AQI 1–4 hours ahead using IQAir API data, meteorological inputs, and seasonal patterns — proactively controlling air purifiers and ERV before PM2.5 reaches hazardous levels, reducing average exposure by 40–60%.

Build a fully local voice assistant on Raspberry Pi 5 using Whisper.cpp (STT), Ollama + LLaMA 3.2 (local LLM), and Piper TTS — comparing privacy, latency, and monthly cost against Google Assistant and Alexa for the Thai market.

Deploy Thai-language Voice AI in Home Assistant using Local Whisper STT and Piper TTS for medication reminders, voice-controlled smart home, and LINE OA emergency SOS — all privacy-first local processing.