Rosales Recommendations

I vibe-coded all of this using Muse, my AI agent. It created all the code. Talk to Your Editor is the working product from the project I described when I applied: an editing platform that learns a creator's judgment from that creator's own finished work. It is a Mac app that edits talking-head videos through conversation. You drop in a raw recording, talk to the editor in plain words, and it makes the edit: taking out retakes and repeated lines, tightening the quiet between your words, and following the directions you give it in the chat. In the chat, you give it a before and an after (your raw video and your finished edit) and it measures what you changed: how much quiet you leave between sentences, whether you keep your breaths, where your first line starts. You can also point it at a reference video whose style you like. And inside the editor, every decision you make teaches it: when you keep or cut a spot it flagged, it proposes one rule in plain words, and saves it only when you say yes. It tells you what it learned, and the next edit uses it, so every edit makes it more accurate. Likely retakes are cut and shown as spots you can bring back with one tap; genuine judgment calls wait as review cards for your call. Nothing is final until you say so: undo restores your exact video from before the edit. You can also fix the edit by hand, right in the transcript. Select words and press Backspace to cut them, or use the toolbar above any selection to cut, bring words back, or comment. Shortened pauses show as small marks between the words; click one to step that pause by a tenth of a second, hear the join, or restore the recorded pause, never longer than the pause you recorded. Undo works on all of it. When the edit is done, the Finish screen frames the video for 9:16 and 16:9 the way your reference frames it, exports the finished video at the loudness platforms expect, and exports the timeline to Final Cut Pro, Premiere, and DaVinci Resolve. I built it because I live this problem daily. I post a video every day, so editing is hours of my day, every day. Most editing apps make you learn their timeline. This one learns you instead. Measured on a real video of mine, Day 30 of my cold plunge series: the app's edit matched my finished edit on 667 of its 692 words, 96.4 percent, in our test environment on the final build. Simply: Day 30 is the video my style pack learned from, and all 12 of its calibration examples come from it, so this score is measured on the video it learned from. It is not a general accuracy number. On a new video there is no score; the test is watching the edit, which is what my demo does. The held-out test is Day 48, a video the app never learned from, and it is in the demo: on a video it never learned from, it kept about 90% of my words, with no dead air, in about 3:17 against my manually cut 2:42. The app carries more than 1,900 automated tests, and every version I test is rehearsed from a clean install first.