One person.A lab that ships every day.

Built By AI Lab runs four products with AI models on its own hardware and Claude as reviewer. Writing, video, quality control and publishing happen every day, for the cost of electricity.

Four products, one pipeline.

Every product runs on the same machinery: local models draft, Claude reviews, a person approves. Pick one to see what it does.

Kairo matches a person's real skills, free hours, budget and hardware to AI income opportunities, then shows the starting cost, the time to a first client and how hard it is.

Open Kairo

How a piece of work gets out the door.

The same five steps for an article, a video or a product page. Each step can stop the line.

Step 1

Draft locally

A 14-billion-parameter model and an image model run on the lab's own GPU, so every extra draft costs nothing.

Step 2

Claude reviews

Claude reads every draft and can rewrite it from scratch when it falls short.

Step 3

Facts and voice checked

Text starts from a dated fact sheet. Voices are compared with the script word by word.

Step 4

A person approves

One message a day on Telegram approves the batch. Nothing publishes on its own.

Step 5

Publish and measure

Each piece goes to its own channel, and a privacy-friendly counter measures real visits without cookies.

Claude does three jobs here.

Local models do the volume. Claude does the work where judgement matters.

Reviewer

Checks every text and script for quality and facts before it reaches a person, with full authority to rewrite.

Builder

Through Claude Code it writes and repairs the tools, websites and automations the whole lab runs on.

Operator

Watches audio, video and publishing results, and stops the line when something looks wrong.

Rules the system runs on.

Each one came from building the lab and is now written into the code.

Stop, never guess

If a settings file can't be read, the system stops and alerts instead of falling back to defaults.

Facts come first

Writing starts only from a verified, dated fact sheet, and won't run if the sheet is out of date.

Draft, review, fact-check

Three separate steps, done by different models, so one mistake can't pass through all three.

IDs are text

Account IDs are stored as text, so a long number can never be rounded onto the wrong account.

Right channel only

Content goes out only on its own channel. The wrong voice or language never ships.

Every word heard

Generated voices are transcribed by two independent recognisers and must match the script before approval.

Small teams deserve this kind of machinery.

Most small businesses can't afford a content team, a QA team and a developer. This lab shows one person can run all three with AI. The next step is turning the pipeline into tools other solo founders can use.