Senior Systems Engineer by title. Platform engineer, AI architect, and fullstack developer by practice. 20+ years turning infrastructure into systems that think, adapt, and ship. Currently orchestrating 30 AI agents across 46 repositories through a custom ops dashboard.
I started in operations — keeping servers alive, managing deployments, watching dashboards at 3 AM. That evolved into infrastructure automation, then into platform engineering, then into building the tools that build the tools.
Today I design interconnected systems where AI agents coordinate development work across an entire ecosystem. 30 specialized agents — each with scoped permissions, custom skills, and independent verification workflows — working in parallel across Python, TypeScript, Rust, C++, Go, and C# codebases, tracked through a custom ops dashboard.
The six projects below are the main line of work. They're the ones I'd defend in a design review: a from-scratch kernel, a local inference platform, a secrets broker, an encrypted messaging stack, an AI game engine, and the pipeline that audits all of it. Everything else lives in Side Quests — bots, games, a Linux distro, home automation, and a few things that didn't pan out.
Anonymized case studies from enterprise engineering work.
Migrated an entire engineering organization's codebase to a new Git org with GitHub Actions deployment pipelines. Built GitHub Apps integrations, implemented HashiCorp Vault with JWT authentication for secrets management, and integrated Artifactory for artifact storage. Automated dozens of operational processes including certificate provisioning via SecretStore.
Built a custom GUI system (Star Trek LCARS-inspired) for the engineering team featuring an integrated RAG-indexed codebase chatbot. Engineers search the entire codebase in natural language. Upgraded legacy Windows Forms to modern WPF, integrated with GitHub Actions and Artifactory. The dashboard became the team's primary interface for build monitoring, deployment tracking, and documentation search.
Six projects, not thirty. These are the ones I'd defend in a design review — the rest are in Side Quests.
Two parallel from-scratch x86_64 kernels — JefeOS (C++) and JefeRust (Rust) — held at near-parity across the core OS, networking, and crypto. Own libc, ELF loader, ring-3 userspace, preemptive scheduler, NTFS read/write, GUI window manager. Full network stack from the E1000 driver up through TLS 1.3, an SSH 2.0 server, SFTP, and authenticated NTS time. POSIX 1003.1-2024 implementation alongside a WSL1-style Linux ABI compatibility layer, now scoping Xylem, a native-clustering direction.
Local inference platform on my own hardware. RAG semantic search over multiple ChromaDB collections, multi-model routing (vLLM → Ollama → hosted fallback) with per-backend timeouts, MLflow experiment tracking, and a 3-tier LLM evaluation harness. Powers the assistant further down this page.
Automated code review across the ecosystem. It started as an LLM-driven sweep — then I measured the output against real code, found the findings weren't reliably grounded, and cut that path. What shipped instead is deterministic: Semgrep SAST plus Trivy, grype, pip-audit and npm-audit, publishing structured findings to an ops dashboard and Discord. Killing the AI feature made the tool trustworthy.
HashiCorp Vault-style secrets broker. Wrap/unwrap workflows, TTL-bound service tokens, rotation automation across 10+ internal services. SQLite-backed with server-side AES-256-GCM and PBKDF2-SHA512 master-key derivation; Express/TypeScript control plane on Titan.
E2E encrypted messaging platform. AES-256-GCM encryption, RSA-4096 key exchange, WebSocket real-time, bot/webhook system. Four clients: React web, Tauri desktop, native Android (Kotlin/Compose), API.
Interactive tabletop RPG campaign platform with AI dungeon master, TTS narration, and AI-generated scene illustrations. Campaign management with persistent world state and character tracking.
The eight pieces everything else plugs into. Inference, secrets, identity, CI, and the dashboard that watches all of it — running on my own hardware.
RAG-powered assistant trained on my professional background. Searches a ChromaDB vector store with semantic retrieval. Try it.
Documented engineering sessions. Each one written by the agents that did the work, with independent security verification.