Back to the work

M4 MediaStudio

I worked on the connections between an episode brief, its edit, its captions, and the material a creator reviews.

Co-development · Full-stack & media engineering
TypeScript · Remotion · FFmpeg

My part in the studio

A recording is only the starting point for a creator’s work. It still needs an editorial direction, an edit, readable captions, related posts, and a decision about what is ready to share. M4 Media Studio brings those stages into a shared production workspace.

I contributed full-stack and media engineering as part of the product’s co-development. My work covered AI brief normalization, review and captioning workflows, and the connection between Remotion rendering and FFmpeg processing. I also co-developed planning, recording, and episode-editing experiences. The common thread was continuity: keeping the creator’s original material and decisions connected as an episode became several different outputs.

The podcast workspace entry point · Cropped from an existing product capture.

The M4 podcast workspace entry point, with a Start new podcast action and planning canvas illustration

The M4 podcast workspace entry point, with a Start new podcast action and planning canvas illustration

The podcast workspace entry point · Cropped from an existing product capture.

Making source material usable

Planning needed to produce something the rest of the application could use. I built normalization around generated episode briefs so a malformed metadata field would not discard an otherwise useful response. The workflow preserves supplied episode titles, premises, angles, and outlines. Missing information becomes an explicit gap or an open question, while proposed structure stays identifiable as a proposal. That distinction gives the creator a concrete brief to revise before recording.

Captioning required another kind of translation. An excerpt begins at a different time from its source recording, so transcript timestamps must move with the edit. My captioning work selects the words within a clip, rebases their timing, and groups them around punctuation, speaker changes, pauses, and reading length. The same cue model can become subtitle files and rendered captions.

I connected that work to the media pipeline. FFmpeg extracts the selected source interval; Remotion renders the final vertical composition with captions and a branded ending when configured. A thumbnail is then taken from the completed render. This makes the preview correspond to the actual delivered asset and carries the brand treatment through production.

Keeping outputs reviewable

I worked on review as a distinct production stage. Generated clip posts retain their episode link, selected time range, transcript excerpt, and media references. They enter the system as drafts, giving a reviewer the context needed to judge an excerpt alongside its source.

Content packages also describe what remains unfinished. Missing captions, disconnected destinations, absent clip media, and skipped generation outputs appear as specific blockers. The package can remain partially blocked while usable work is available for review. Reviewers can submit, approve, or return posts to draft, and scheduling requires a destination.

These connections are the part of M4 I wanted to strengthen: a production process where generation produces usable material, and each subsequent step gives the creator enough context to make the next decision.