Deepa Shenoy explains: Brutalist Video Submission & QC Process

This is Deepa Shenoy's week four progress update with Humanitarians AI, narrated by Liam on her behalf. By this point each video exists in two native versions and goes through human review, technical checks and a careful submission process. The video walks through those stages and ends with a point about what finishing a media project really involves.

Two native versions of every video

Each video now has a 16 by 9 landscape cut and a 9 by 16 portrait cut. The portrait version is composed independently for the taller frame. It is not simply the landscape image trimmed down, so the layout is designed for the shape it will be seen in.

Two human review gates

The pair of cuts moves through two human review gates. First she listens to and approves the narration before any video rendering takes place. Then, once both complete cuts are ready, she watches them before approving the finished pair. The order matters: approving narration first means the render is not spent on a script that has not been signed off.

Automated checks that support review

Automation helps along the way. Depending on the project, the checks can include output resolution, frames and layout, typography, safe areas, audio and loudness, complete decoding and master file properties. The video is clear about their role. These checks support review and do not replace it.

A clear place for every file

She has also been building an organized local submission folder with a separate folder for each content piece. Inside each one, a video folder holds the approved MP4, and a GitHub folder holds the markdown documentation along with the landscape and portrait beat sheets. That separation gives each file a clear place and keeps video masters out of the documentation folder. The beat sheet stays with the written project material, where it describes the planned beats for each format.

Before calling a copy ready, she checks that:

• There are two MP4s, one landscape and one portrait.
• Both beat sheets and the relevant documentation are present.
• The GitHub folder contains no video files.

Checking file integrity with hashes

For file integrity, she compared SHA-256 hashes on the latest copied video pair. A file hash works like a fingerprint, and matching values help confirm that the copy matches its approved source. Doing this for the latest pair made the verification step concrete. She checked the names and counts in each folder, confirmed both beat sheets, and matched each copied MP4 hash against the source.

The broader lesson

Finishing a media project takes more than generating pictures and sound. It also takes review, quality control, clear organization, useful documentation, and checks that the right files ended up in the right places. At the time of the video she was preparing and organizing the approved materials for submission, and had not uploaded or published them. This step was about getting the local package structured and verified.

Key takeaways

• Make each video in two native versions, 16 by 9 and 9 by 16, with the portrait cut composed on its own.
• Approve the narration before rendering, and approve the finished pair of cuts after watching them.
• Automated checks cover things like resolution, safe areas and loudness, but they support human review rather than replacing it.
• Give each content piece its own folder, with the MP4 in a video folder and documentation in a GitHub folder.
• Keep video files out of the documentation folder.
• Compare SHA-256 hashes between each copy and its approved source.
• Finishing a project includes review, organization and verification, not only generation.

Try it yourself

Take one finished video of your own and run the same checklist on its submission folder. Count the MP4s, confirm the beat sheets and documentation are present, make sure no video files sit in the documentation folder, and compare file hashes between the copy and its approved source.