I built a fully local AI voice assistant that combines wake-word detection, speech-to-text, a large language model, and text-to-speech into a single offline system. The assistant listens for a custom wake word, transcribes spoken input, processes the request through a local language model, and responds using synthesized speech.
The system is designed to run without relying on external servers, which improves privacy, reduces network dependency, and gives greater control over the full inference pipeline. Each component is connected into a real-time interaction loop: wake-word activation triggers speech recognition, the transcribed text is passed to the language model for response generation, and the final output is converted back into natural-sounding speech.
This project demonstrates my ability to integrate multiple machine learning components into a complete applied AI system. It also highlights experience with local inference, voice interfaces, natural language processing, and end-to-end system design.