Case study

YouTube Video Automation

From a topic file to a publish-ready package: a human-approved YouTube automation producing narration, storyboards, captions, thumbnails, and Shorts.

Discuss a similar product

Platforms

Automation

Role

Architecture and end-to-end development of the pipeline.

Tech stack

Node.js · FFmpeg · OpenCLIP

Status

Launch-ready

Problem

What was needed?

Publishing explainer videos regularly meant an unsustainable manual chain of research, narration, editing, captions, thumbnails, and uploads.

Solution

How was it solved?

A single-command pipeline: narration and storyboard are generated from an editorial brief, stock footage is scored against the script by a local OpenCLIP model, Whisper produces word-timed captions, Shorts are derived from the long cut — and nothing reaches publish without passing a hard quality gate.

Outcomes

What came out of it?

Visual-script alignment via local OpenCLIP scoring, no cloud vision cost

Quality report with LUFS/true-peak checks + mandatory human approval

SQLite institutional memory preventing topic and footage repetition

Automatic Shorts derivation and timezone-aware scheduling

Build note

Build narrative

The pipeline has shipped 9 long-form videos and 36 Shorts for a YouTube channel. Its defining trait is discipline rather than autonomy: scripts stay editorially human-made, the system accelerates production, and publishing requires a passing quality report, a selected thumbnail, and explicit human approval together. Built-in cost accounting reports per-video token, character, and infrastructure spend.

Product experience

Client app and management panel

One tracking system, delivered through two interfaces shaped around each role’s daily workflow.

Discuss a similar product