logSguarDian is a graduation thesis project at Universidad del Valle de Guatemala (2026) — built not just to be defended, but to be a real npm package a small team could put in front of production traffic.
Cybersecurity integration, dataset verification, library architecture and structure, testing and core logic.
AI/ML model integration, feature engineering and model training, library architecture and structure, testing and core logic.
Published: logsguardian@0.1.0 is live on npm. Detection, the CLI, and the evaluation suite are functionally complete. The repository is in a documentation and packaging polish pass rather than active feature development.
The model versions currently shipping are rf_v10 / if_v9, running on a worker-pool architecture (1 dedicated RF worker, a round-robin pool of 2 IF workers). 68/68 extractor tests, 147/147 core middleware tests, and 113/113 CLI tests pass. A handful of middleware tests depend on real setTimeout margins and can flake under full parallel CI load — always pass individually, tracked as a known CI quirk rather than a product defect.
This is a research project with real, stated limitations — not a finished commercial product. If you run WAFs or RASP tooling in production, or want to poke holes in the methodology, that's exactly the feedback that's useful. Open an issue, or read what we already know is imperfect before reporting a duplicate.
github.com/xtsebas/logSguarDian — pnpm monorepo, MIT licensed.
MIT — see LICENSE in the repo.