RVP SLED Central @ Elastic. Building AI agent fleets on self-hosted hardware since before it was cool.
My public repos are the visible edge of a homelab AI fleet — multiple servers, dozens of agent profiles, local LLM serving, and memory systems that learn while I sleep. The philosophy: own your inference, measure everything, automate the maintenance, and publish what it teaches you.
| Repo | What it is |
|---|---|
| praxis | ⭐ The hub — hard-won lessons from running the fleet: serving recipes with verification gates, agent memory architecture, benchmark methodology, curated links |
| mnemograph | The in-process knowledge graph for AI agents — no external databases, no network calls, zero infrastructure |
| focus | Self-hosted web launchpad — Flask + vanilla JS, the start page every browser in the fleet calls home. Inspired by Bonjourr. |
| Clearvoice | Privacy-first meeting assistant for Linux — record, transcribe, diarize, and summarize meetings with local AI. The fleet's meeting-intelligence frontend. |
| BonjourrNue | Minimalist startpage (upstream fork I track) |
- mnemograph — knowledge-graph memory as an embeddable library
- praxis — the operating education: what deploying all of the above taught us
- Everything else (Agent Zero dashboards, startpages, tooling) exists to keep the fleet running
- build.nvidia.com — where the DGX Spark practitioner community trades notes (find my verification posts there)