1 /** 2 * Joseph Rosenbaum 3 * Systems Designer & 4 * Full Stack AI Engineer 5 */ 6 7 const profile = { 8 name: "Joseph Rosenbaum", 9 base: "Dallas, TX", 10 years: 15, 11 12 stack: [ 13 "Node.js/TypeScript", 14 "Proxmox/Docker/LXC", 15 "LLM Agents + MCP", 16 "M365/Graph API", 17 "PowerShell/.NET", 18 "Caddy/Cloudflare", 19 ], 20 21 certifications: [ 22 "CompTIA A+", 23 ], 24 25 education: { 26 school: "UT Dallas", 27 focus: "Computer Science", 28 }, 29 30 solve(problem) { 31 return this.automate( 32 this.document(problem) 33 ); 34 } 35 };
I'm Joseph Rosenbaum — a Systems Architect and AI Engineer based in Dallas, TX building production AI infrastructure, self-hosted inference, and operator-facing automation for MSP operations.
I built Manifold, a Go LLM gateway routing org-wide AI traffic with priority queues, fallback chains, and full telemetry; RC-AT, a .NET 10 WPF desktop app plus server that automates RingCentral-to-Autotask ticket intake; and 7 MCP servers that expose the MSP stack to AI agents.
I also run a 55+ model local inference stack on AMD Ryzen AI MAX+ 395 hardware with active Unsloth QLoRA fine-tuning, manage 25+ VMs on Proxmox, and ship tools like Conduit, SSHubble, stack-dashboard, and Pulse into real production workflows.
Production LLM gateway routing org-wide AI traffic with OpenAI-compatible APIs, priority queues, capability aliases, fallback chains, SQLite telemetry, and Prometheus metrics for live observability.
Built 7 Model Context Protocol servers for Autotask, Datto RMM, IT Glue, SaaS Alerts, Pax8, M365 Graph, and OneDev, giving AI agents read/write access across the MSP stack.
Self-hosted inference and fine-tuning platform on AMD Ryzen AI MAX+ 395 with 55+ models, benchmark-driven backend tuning, Gemma 4 multimodal support, and QLoRA workflows for MSP domain data.
Multi-tenant Microsoft 365 automation platform with token broker, GDAP lifecycle management, bulk admin consent, async job queueing, and durable handling of MFA and dynamic SPA auth flows.
Approval-gated SSH reverse-tunnel relay and MCP server for AI-agent remote support sessions with ephemeral keys, ForceCommand restrictions, NetBird provisioning, and full audit transcripts.
Production WPF desktop app plus server that turns RingCentral Phone activity into Autotask tickets with semantic embedding classification, technician style profiles, and JSONL correction logging for fine-tuning.
Operational MSP dashboard with dual-model AI chat, 23 tools, and live data gathered in parallel from Datto RMM, Autotask, IT Glue, SaaS Alerts, Microsoft 365, Pax8, and RingCentral.
Windows tray app and dashboard that infers work context from Git state, window focus, and Claude session JSONL logs, enforces cadence, and runs end-of-day reflection against the local inference stack.
Whether you need an infrastructure architect who builds AI-integrated tooling, someone who can own an entire automation practice end-to-end, or a developer who ships real products to production — I'd love to connect.