Due to NDA and to reflect modern generative workflows, the original product logic and ML pipeline have been adapted into an updated interface.

Editing a sports vlog means hours of scrubbing through raw footage, looking for highlights and cutting audio by hand. We built a mobile-first video editor that does the first cut automatically. The user describes what they need, and the app analyzes the footage library (frames and audio), transcribes speech and assembles a vertical video with titles, descriptions and cover frames. I was Lead Designer and worked on the editor and the AI module.

Empty timeline: Replace hours of manual search and cutting with an auto-assembled first draft that the user only refines.

Predictable AI input: The system understood keyword templates, not free text. The task was to make this limitation feel natural, so the user gets the result they expect from a short request.

Editing through text: Link the transcript and the timeline in both directions: deleting a word or a pause in the text trims the matching video and audio.

Complex editor on a small screen: Fit a multi-layer timeline into a phone with a magnetic timeline and gesture controls.

Trust in automatic results: Show how confident the system is in speech recognition and give the user a quick manual fix that doesn't interrupt editing.

Highlight detection: Selection of intense moments (saves, goals, reactions) by audio peaks and motion tracking, with dead time removed.

Template-based AI requests: Keywords map to highlight types and edit styles, so the result is predictable and easy to repeat.

Transcript-based editing: Trimming pauses and filler words directly from the text.

Multi-layer timeline: A main video strip with separate layers for voiceover, sound effects and captions.

Publishing flow: Generation of hook titles, descriptions and cover frames right before export.