I built my own bass practice tool, and I finished it.

Practicing bass well takes two things most tools give you only one of: something to play against, and an honest read on how it went. Backing tracks are static: the same loop every time, in the same key, at the same tempo. And nothing tells you what to work on next except your own memory of a session you were busy playing through.

So I built Bass Face for myself, as a real React product rather than a weekend throwaway. Building for an audience of one is a strange kind of freedom: there was no spec to satisfy and no deadline to cut against, which meant the only thing deciding when it was “done” was whether I actually wanted to open it every day. That's a higher bar than most briefs set.

Bass Face leans on two live services, each doing something a static app can't. Tone.js synthesizes the backing tracks in the browser in real time. This is generated audio, not a file being played back, so the accompaniment can shift instead of looping. Audio is always user-initiated: you press play. That's both the natural interaction and the right accessibility default: nothing blasts out of the speakers the moment the page loads.

The second service is the Claude API, wired in for reflective feedback, turning a practice session into a short, useful read on what to work on next. The interesting engineering isn't either service on its own; it's making real-time audio synthesis and an LLM live together in one coherent interface, with the latency, state, and failure modes of both handled gracefully. That integration work is exactly what product engineering is.

Bass Face covers a lot of distance for a personal project: live audio synthesis on one side, an LLM integration on the other, and a real interface holding both together. That range (comfortable wiring up DSP-style audio one moment and a language model the next) is the part I'm proud of. It's the same instinct that lets me move across a stack instead of staying in one lane.

And because I built it for myself, nothing forced me to cut the last ten percent: the empty states, the play/stop interaction, the moment when there's nothing to reflect on yet. That's the same finish I bring to the products I ship for teams. A personal tool is just the clearest place to show it, because there's no one else to blame for the parts that aren't done.