AutoFix AI

Inspiration

I was curious whether an AI could do more than just generate code once. In real development, most time is spent running tests, reading errors, and fixing things. I wanted to see if an AI could handle that loop on its own.

What it does

I learned how to use Gemini 3 as part of an iterative workflow instead of a single-step generator. I also learned how important clear prompts and error feedback are when building autonomous systems.

How we built it

The project is built in Node.js. A task is given in plain English, and Gemini 3 Flash Preview generates the code and test cases. The tests are run locally. If they fail, the error output is sent back to Gemini 3, which fixes the code and retries.

Challenges we ran into

The main challenge was handling inconsistent outputs and making the retry loop stable. Designing prompts that worked reliably across multiple attempts took some trial and error.

Built With

• chalk
• dotevn
• javascript
• node.js
• ora