Log Detective

Trying to improve RPM packaging experience by analyzing build logs and explaining the failure in simple words.

Welcome to the new experimental functionality of Log Detective, thank you for your interest!

This is our initial prototype that will 'Explain' a log of your choice to you. It has several limitations:

The inference is slow and serial. We can only process a single request in the background. It will take at least 30 seconds to give you a response. In case of multiple requests, it can ramp up to minutes.
We use a general-purpose mistral model in the background. The collected data are not being used here just yet. We are still working on fine-tuning our own model.
Please report any issues you'll encounter. We don't have any alerting in place, the deployment is highly experimental.

About the project

Debugging failed builds is hard

Each build produces thousands of lines of output split among multiple log files. And the relevant error message can be anywhere. It's just like a needle in a haystack.

Does it matter?

Veteran packagers have an intuition where the error message will most likely be, but the process is tedious regardless. Newbies are often overwhelmed by the complexity and miss the error message completely.

What is our goal?

Training an AI model to understand RPM build logs and explain the failure in simple words, with recommendations how to fix it. You won't need to open the logs at all.