AI Engineer · Systems Engineer · Infrastructure & IoT
I build the systems underneath the software — industrial data pipelines, digital twins, edge AI, and the infrastructure that keeps them running.
I'm Roel (yeems214), an AI Engineer at Cubeworks Technology Consulting & Solutions Inc. in Cebu City, where I design and ship industrial IoT, computer vision, and digital-twin systems for enterprise clients.
I think of myself as a systems engineer first. I've been building and breaking machines since I was nine — virtualization, Hackintoshing, OpenCore, hardware reverse engineering — and that habit of understanding the full stack from silicon to service is what I bring to every project.
Education: BS Information Technology (University of Cebu – Banilad) and BS Software Engineering (Lithan Academy, Singapore), completed concurrently, 2026.
| Area | Highlights |
|---|---|
| Industrial IoT & Data Backbones | IEC 104-to-MQTT converter; MQTT/NATS backbone for a 7-site OPC-UA deployment; ReAct-style IoT agent pipelines |
| Computer Vision at the Edge | YOLO-based intake and weighing system with OCR and forklift detection, deployed on NVIDIA Jetson Orin Nano |
| Digital Twins & Real-Time 3D | Three.js twin viewports with live MQTT binding, custom shaders, and gamepad input; neural rendering PoC (G-buffer-conditioned U-Net, Core ML inference on Apple Silicon) with WebGPU integration in progress |
| Agentic & LLM Pipelines | GitHub Actions + LLM pipeline for automated weekly project reporting; domain-specific grammars (LICML, PBML, QAML, MNML, ACML) extending a composable modelling stack |
| Apple Platform | Project-management app with Siri App Intents integration on the latest iOS/macOS betas, including full UAT sign-off |
| Telecom Integration | Multi-tenant SMS gateway on Yeastar TG200 hardware |
Languages
AI / ML
IoT & Messaging
3D & Graphics
Backend & Web
Infrastructure & DevOps
| Project | Description | Stack |
|---|---|---|
| Neural Twin Rendering | DLSS-style G-buffer completion: a compact deterministic U-Net trained on Blender Eevee→Cycles pairs, running via Core ML on Apple Silicon. Phase 2 targets per-twin models with WebGPU inference in the browser. | PyTorch · Core ML · Blender · Three.js |
| Homelab (Proxmox + Tailscale) | Self-hosted Proxmox VE node with ZFS storage, LXC/VM workloads, a thin-client quorum/monitoring device running Uptime Kuma, UPS-backed and reachable over a Tailscale mesh. | Proxmox · ZFS · Tailscale · Debian |
| Scroll-Driven Three.js Visualizations | Series of interactive 3D narrative pieces — datacenter walkthrough, warp-drive explainer, environmental dioramas — plus a digital-twin demo rig with Gerstner wave shaders, gamepad control, and MQTT binding. | Three.js · GLSL · MQTT |
| yeems214.xyz | Self-hosted platform for personal web apps and services. | Docker · Nginx |
Earlier work: ShareX Image Server (Spring Boot / MySQL) and its React Native companion Duffin's Image Drops; Diana AI, a Gemini-powered chatbot for music enthusiasts; Duffin's Music Player, a Java WAV player with time-synced lyrics.
Open to conversations about industrial IoT, edge AI, digital twins, and self-hosted infrastructure.