When we decided to bring a slate of Hindi microdramas to Indonesia, everyone in the room wanted to talk about translation. Would the jokes land in Bahasa? Would the names sound right in a mouth they were never written for? Those are real questions. They are also, it turns out, the easy twenty percent.

A microdrama is sixty or eighty episodes of ninety seconds each. Multiply that across a catalogue and you are no longer translating a script, you are running a factory. The script is the raw material. The product is the pipeline: casting voice actors who can carry an emotion that was written in a language they do not speak, briefing them on a story culture they did not grow up inside, recording, doing quality control at a volume where you physically cannot listen to everything, and shipping on a calendar that does not care that your best actor has a cold.

You are not translating a script. You are running a factory, and the script is only the raw material.

The eighty percent nobody demos

Here is where the name of this site earns its keep. At our volume I cannot personally review every line, so we built a scoring rubric: does the emotion in the take match the emotion in the scene, is the name pronounced consistently across every file it appears in, does the mix sit where it should. A rubric lets a small team hold a large catalogue to one standard. It also has a margin of error, and some of its cells lie confidently. A take can score well on every measurable axis and still feel dead, and the reviewer who catches that is worth more than the rubric that missed it.

So the real work is not writing the rubric. It is knowing where to distrust it, and building the operation so a human lands on exactly the takes the score cannot be trusted on. Most of my job is designing that routing.

Where the machine actually helped

This is also the point where AI stopped being a demo and started being a coworker. Not for the creative call, machines are still bad at deciding whether a performance moved you, but for the operational eighty percent: flagging takes where the vocal emotion drifts from the scene, catching one mispronounced name across four hundred files in the time it takes to make coffee, drafting a first-pass rubric that a human then argues with. It is not judgement. It is leverage on the part of the job that was always going to be tedious, which frees the judgement for the part that was always going to be hard.

The lesson I keep relearning is that the interesting problem is almost never the one everyone walks in talking about. The translation is the headline. The operation is the product. Get the second one wrong and the best translation in the world still ships late, off-brand, and unwatched.