In a recent blog post, Eugene Yan details his hands-on experiment to build an AI coach aimed at helping manage his own "monkey mind." The system brings together four core components: speech-to-text to capture the user's spoken input, a large language model to generate thoughtful coaching replies, text-to-speech to deliver those replies audibly, and a virtual phone number to make the whole setup accessible through ordinary calls.
Yan's write-up is a practical walkthrough rather than a theoretical overview. He explains how these building blocks integrate, and while the post is a single source, it offers a concrete example of assembling off-the-shelf AI tools into a voice-driven coaching assistant. The emphasis is on the architecture and the user experience of talking to a coach rather than typing to one.
The project highlights a trend in personal AI tools: moving beyond text-based chatbots to voice-native interfaces. By using a virtual number, the coach can be reached from any phone, which lowers the barrier to regular use. Yan does not compare his approach to others or discuss alternatives; the article focuses solely on his design and implementation choices.