AI-powered mix feedback

Why I built this

Getting feedback on a mix usually means waiting for someone else to listen - a friend, a mentor, or a paid service. I wanted a tool that could give instant, structured feedback on a mix, focusing on the technical aspects that are easy to miss when you've been listening to the same track for hours.

How it works

Users upload an audio file, and MixCrit processes it through a Web Audio API pipeline to extract frequency data, dynamic range, and stereo information. That analysis is sent to Groq's API with a specialized prompt that returns feedback structured by category - EQ, dynamics, stereo image, and overall balance.

Interesting decisions

The biggest challenge was making the AI feedback actually useful. Early prompts returned generic advice - "your low end could be tighter" - regardless of the input. I iterated on the prompt engineering to include the raw data, which forced the model to ground its feedback in the actual audio rather than guessing.

I also chose to process the audio entirely client-side using the Web Audio API rather than uploading files to a server. This keeps the app fast and avoids storing user audio, which felt important for a tool aimed at unreleased music.