Skip to content
NORTH
Book a call
All work

Linty

AI code review that writes the fix, writes the tests, and runs them against both versions.

Kind
Hackathon · AI code review — IBM Bob hackathon
Year
2026
Role
Product design, UI design, Build
Built with
Next.js, Tailwind CSS, IBM watsonx.ai, IBM Bob

The growth problem

Any model can produce a list of things that might be wrong with your code. A suggestion costs nothing to generate and nothing to be wrong about, so it earns about as much trust as a stranger's guess.

What the design does

  1. Positioning

    Sells verification rather than detection. Four agents run in order — analyse, fix, test, report — and the tests the third one writes are then executed against the broken code and against the patched code. The output stops being "this looks like a bug" and becomes "here is a test that failed before and passes now", which is a different class of statement and the only one worth acting on.

  2. The path

    Ships recorded runs instead of demanding credentials. Live analysis needs watsonx keys on the server, so rather than a button that fails for every visitor, the deployment carries replayable runs: the issues found, the fixed code side by side, the generated tests and whether they pass. The limit is written on the page. Someone with two minutes sees the whole pipeline work without signing up for anything.

  3. The feeling

    An IDE register — near-black ground, monospace paths, the agent pipeline laid out as four named stages that light as they run. It looks like a tool rather than a landing page, which is the right register for something asking to touch your repository.

Try it here

The live product, running in the page.

linty.vercel.appOpen
Guild