LLM security research
BadToken HYDRA
Automated LLM red-teaming framework built around one scheduler and one evidence format.
PythongarakPyRITvLLM
In development
All projects
Case study / Infrastructure
Securityhomelab
TrueNAS and ZFS infrastructure that doubles as an offline target for red-team work.
Overview
Why run a lab
The homelab is where I get to be the administrator, the attacker, and the person on call. It gives the research an offline target and gives me real infrastructure to break and fix.
- TrueNAS / ZFSStorage and app platform
- Local LLM containerOffline red-team target
- Pi-hole + UnboundDNS filtering, Docker Compose
- Nextcloud, JellyfinSelf-hosted services
- Uptime monitoringAlerts to my phone
- Monitoring toolPython, SQLite, Trivy
Walkthrough
A six-minute tour
Lessons
Things that broke
- Recovering after a move. A subnet change and an interface misconfiguration took the server offline; I brought it back, confirmed ZFS pool health, and set a DHCP reservation so it wouldn't drift again.
- Stale DNS. When the server's IP shifted, a stale Pi-hole DNS reservation broke name resolution until I traced it back.
- Security monitoring tool. A Python tool with SQLite storage and Trivy scanning, built in phases. It isn't on GitHub yet.
Roadmap
What's next
Wazuh is the highest-leverage addition, since it brings SIEM-style detection home. Also planned: Tailscale, Vaultwarden, and Nginx Proxy Manager.