CLIENT

DICTA (AI)

CONTEXT

I TYPE A LOT EVERY DAY, AND AT SOME POINT, IT STARTED TO FEEL LIKE A WASTE OF TIME. THE OBVIOUS FIX WAS A DICTATION APP, SO I TRIED THE POPULAR ONES. THEY WORKED, BUT THEY ALL WANTED A MONTHLY SUBSCRIPTION I'D PAY FOREVER, THEY ALL SENT MY AUDIO TO SOMEONE ELSE'S SERVER, AND NONE OF THEM LET ME CHANGE THE ONE THING I WANTED TO CHANGE. SO I BUILT DICTA.

IT LIVES IN THE MACOS MENU BAR: HOLD A KEY, TALK, LET GO, AND THE TEXT APPEARS WHEREVER YOUR CURSOR IS — SLACK, EMAIL, FIGMA, AN EDITOR, ANYTHING.

WHAT I DID

DESIGNED AND SHIPPED A NATIVE MACOS APP END TO END — INTERACTION MODEL, FLOATING HUD, BRAND, ICON, ONBOARDING, AND SETTINGS. BUILT IN SWIFT WITH CLAUDE CODE; I OWNED THE PRODUCT DECISIONS AND DIRECTED IMPLEMENTATION.

THE CHALLENGE WASN’T TRANSCRIPTION. IT WAS MAKING AI FEEL NATIVE TO THE OS. A HOLD-TO-TALK KEY THAT DOESN’T BREAK. A HUD THAT NEVER STEALS FOCUS. ESC TO CANCEL. AND TEXT THAT WORKS IN ANY APP WITHOUT INTEGRATIONS.

THE DECISION

NEITHER SPEECH ENGINE WAS ENOUGH ALONE. WHISPER IS MORE ACCURATE, ESPECIALLY IN SPANISH, BUT HAS NO LIVE TRANSCRIPTION. APPLE IS INSTANT AND SHOWS LIVE TEXT, BUT MAKES MORE MISTAKES.

SO I USED BOTH: APPLE SHOWS LIVE TEXT WHILE YOU TALK, THEN WHISPER REPLACES IT WITH THE MORE ACCURATE VERSION ON RELEASE.

THE TRADE-OFF: TEXT CAN CHANGE AT THE END. I KEPT IT BECAUSE SILENCE FEELS WORSE THAN CORRECTION. NEXT: MAKE THE HANDOFF FEEL LIKE A CORRECTION, NOT A GLITCH.

AA / ALL RIGHTS RESERVED
13:07 GMT-5
LIMA, PERU
AA
13:07 GMT-5
LIMA, PERU
13:07 GMT-5
LIMA, PERU

Create a free website with Framer, the website builder loved by startups, designers and agencies.