Designed and built a self-owned small-scale AI and compilation infrastructure to eliminate managed-cloud lock-in, data egress costs, and opaque token billing.

Architected a multi-node build cluster featuring Slurm workload scheduler, distributed sccache compilation with multi-level Redis caching, and per-node cgroup memory guards for Rust, C++, and Go projects.

Designed and open-sourced kepr, a per-user credential vault in Rust utilizing AES-256-GCM encryption, Argon2id master-key derivation, RAM-sealed memfd secret delivery (no plaintext in /proc/environ), and hash-chained audit trails.

Built local LLM routing tiers and an OpenAI-compatible gateway so workstation and cluster resolve models through self-owned hardware. Maintained cross-node parity via deterministic dev-setup and drift detection tooling.