[GSoC 2025] Finalizing Proposal for Fedora AI Log Triage System – Feedback Welcome!

Ask Fedora
1

Hi Fedora mentors and community,

I hope you’re all doing well. As I near the GSoC 2025 application deadline, I wanted to share a final version of my proposal for the AI-Powered Log Triage and Security Alert Aggregator project.

The project aims to:

  • Streamline Fedora’s log analysis using AI/ML (via OpenAI & Hugging Face)
  • Build a secure, extensible alert triage dashboard
  • Improve incident response time and signal-to-noise ratio in logs

Proposal PDF: GSOC Project Proposal.pdf - Google Drive
Let me know if there’s anything that needs fine-tuning before submission. I’d love to make the most of this opportunity and am committed to contributing regardless of the outcome.

Thanks again for your support and guidance!

Cheers,
Justin Wilson John

2

I need a login to access the g-drive link, which I do not have.

Is there a reason you chose to use OpenAI? I would have preferred to see a more open model if there is a long term plan to integrate with Fedora.

3

Yes, sorry for that. Here you go: GSOC Project Proposal.pdf - Google Drive

You’re right — using OpenAI was an area also concerning me while creating the proposal. My initial idea was to use open, local models from HuggingFace or Ollama wherever possible. I’ve already started exploring models served via Ollama like:

  1. Phi-2 – small and efficient for general reasoning

  2. DistilRoBERTa / DeBERTa – great for classification and intent recognition

These models can be fine-tuned or prompt-engineered to support the log triage and alert classification use case.

Thanks for pointing it out, @theprogram

4

Really cool that your on to it Justin,
It will be great to get some local help deciphering journalctl logs!