project · February 08, 2026 · 3 min read
Content Generation Harness
A multi-agent harness I built that turns a single topic into a finished short-form educational video, and cut my video-making time by about 10x.
Content Generation Harness
From guest-lecturing and reviewing student projects, I noticed the same thing over and over: students absorb core CS concepts much better in short, bite-sized form. I wanted to publish short educational videos regularly, but making them by hand ate up all my time. So I built a harness to do most of the work for me.
It is a Claude-Code-style setup, a little ecosystem of scripts, CLI tools, and plugins that takes a topic and produces a finished video. Give it something like “Database Replication” and it will:
- source good reference material on the topic,
- draft a script tuned to the audience metrics I care about,
- run that script through a review gate where a few sub-agents score it and keep refining until the average score hits 8,
- use two more sub-agents to decide which scene is best rendered in which tool (Remotion or Manim),
- render the scenes, generate the narration with a TTS service, and stitch everything into the final cut.
I also built a small video-reviewer web app for the human-in-the-loop part. I can click a specific pixel at a specific timestamp and leave a change note, and the harness reads those notes back in and acts on them. All of this cut my video-generation time by at least 10x, and I actually started publishing content instead of just meaning to.
The two repos below are the pieces of it: eduvid is the engine (the CLI tools
and binaries that handle TTS, rendering, and compositing), and techniches is
the layer of plugins and skills that orchestrates the whole pipeline on top.
