Prosodia: An Audio Jev
Building the first audio-native System One model, and the results I found, then un-found when I actually controlled for them.
2461 words · 12 min · 2026-09-21
TinyAya Expedition: S2S Translation from a Text Backbone?
A retrospective on building a Turkish-Hindi speech-to-speech translation model from scratch — the decisions we made, the ones we got wrong, and what we would do differently.
1784 words · 9 min · 2026-07-10