class SantiagoR:
location = "Medellín, Colombia 🇨🇴"
company = "VIA TERRESTRE S.A. — Envigado, Antioquia"
role = "Sole Engineer · Systems Architect"
current = "Plataforma VIA — production ERP, built solo from scratch"
infra = ["Ubuntu 24.04", "OPNsense", "Docker", "Tailscale", "Cloudflare", "2× RTX 5090"]
side_quest = "Lucio & Co — pet products in Fusion 360,🐾"
workflow = "Architect ↔ Executor — two Claude Code agents, I make the calls"
hobbies = ["FPV Drones 🚁", "Betaflight PID tuning", "Self-hosted everything"]
def say_hi(self):
print("I build real systems under real regulatory constraints.")
print("No managed databases. No Alembic. No shortcuts.")A production, domain-driven ERP replacing a decade of Excel, paper signatures, and fragmented Google Drive for a Ministerio de Transporte-regulated passenger-transport operator.
Domains: Human Capital · Fleet & Maintenance · Operations · HSE/LMS · Documental Compliance · AP/Settlements · Accounting · Alerts
Integrations: SIIGO Nube API · DIAN e-invoices · ONLYOFFICE · Grandstream UCM6300A PBX · Cloudflare Zero Trust · Huawei E3131 GSM OTP
🐾 Side quest — Lucio & Co · customizable pet products, designed around each pet — two Claude Code agents drive Fusion 360 through MCP, and I make the design calls.
The AI tier of Plataforma VIA, and the project I'm learning the most from right now. I run local inference on dual RTX 5090s — from kernel-module driver fixes for Blackwell, through the Docker GPU runtime, to a Graph-RAG pipeline over the company's regulated docs vault. This data can't leave the building, so instead of cloud APIs I'm building the on-prem alternative — and documenting every step like infrastructure, not experiments.
Stack: Ollama · qwen3:32b · bge-m3 embeddings · LightRAG · GraphRAG · RAG-Anything (multimodal) · Docker GPU runtime
Status: inference is up and benchmarked; now wiring the RAG pilot into the platform's AI copilot endpoints — and re-learning half of it every week, because this field doesn't sit still.
🤝 If you're working on something similar — local LLMs, RAG over messy real-world documents, GPU homelabs — I'd genuinely love to compare notes. Reach out.
⚙️ Backend
🖥️ Frontend
📱 Mobile
🚢 DevOps & Infrastructure
📐 CAD & Product Design
🧠 AI & LLM Ops
🔧 Tools & Environment
| 🚁 FPV Drones | 🖥️ Self-Hosted Infra | 🤖 AI Exploration | 🐾 Product Design |
|---|---|---|---|
| Freestyle & long-range builds | OPNsense · Tailscale mesh | Claude Code power user | Lucio & Co — pet products |
| Betaflight PID tuning & flash | Asustor NAS · Grandstream PBX | On-prem Ollama · LightRAG · GraphRAG | Fusion 360 parametric CAD |
| FPV stack wiring & soldering | Nightly pg_dump · RAID 5 NAS | Fusion - CAD | Custom designs, fit to each pet |