Ygor Lemos ygbr

🧭 whoami

$ whoami
ygor · CTO & Co-Founder @ eles · systems architect · infrastructure operator

$ uptime
 20+ years in production · load average: high · downtime tolerance: 0.00

I build and operate systems where failure is expensive: carrier-grade telephony, a consumer platform that scaled past 4 million users, and now healthcare infrastructure handling prescriptions, pharmacy operations and sensitive patient data across Brazil.

Today I'm CTO & Co-Founder at eles (@eles-tech), a Brazilian digital health company building platforms for men's and women's health. I own the entire technical surface: architecture, cloud, backend, frontend, mobile, ML inference, CI/CD, and the infrastructure-as-code that holds it all together. CTO is the title; the terminal is where I actually live.

🛤️ The Road So Far

	Era	Chapter
☎️	2003 → 2011	Systems and network architecture in telecom. Carrier-grade VoIP with Asterisk, SIP, IAX and SS7 signaling: real-time systems where a dropped packet is a dropped phone call. Linux, UNIX and BSD administration at scale, long before "DevOps" had a name.
📱	2012 → 2024	Co-founded PiniOn and ran engineering as CTO for ~12 years. Built the largest crowdsourcing mobile app in Latin America to 4M+ users, and wrote the early platform solo: backend, apps, infra, all of it. Recognized as an Appcelerator Titanium Titan along the way, later migrating the mobile world to SwiftUI and Kotlin/Jetpack Compose.
🏥	2024 → now	Co-founded eles. Digital health, AI-native from day one: pharmacy management, medical prescription review, operations dashboards and fiscal integrations, all engineered around LGPD data sovereignty.

⚙️ What I Run Today

The production stack at eles, designed, built and operated hands-on:

🐍 Python 3.14 · FastAPI · SQLModel on Google Cloud Run (southamerica-east1), strict typing, mypy strict, 100% IaC
⚡ React + Vite served from the Cloudflare edge
🏗️ Pulumi for every last resource. If it exists in the cloud, it exists in code.
🧠 Self-hosted medical LLMs with vLLM, TensorRT-LLM and ONNX on NVIDIA L4 (inference) and B200 (training)
🔒 LGPD data sovereignty as an architectural constraint, not a compliance checkbox: patient data never leaves Brazil
🧾 Mission-critical fiscal pipelines (NF-e, NFS-e, SEFAZ) where one rejected document can stop a pharmacy from selling

🤖 AI-Native Engineering

I rebuilt the way software gets shipped here, and I have the receipts:

🚁 Fleets of coding agents running in parallel: Claude Code, Jules and Codex
📐 AI Spec Driven Development: I write the architecture and the specs, agents write the code, nothing merges without my review
🚀 Three internal production systems shipped in under two months with this setup
🎯 The bottleneck moved from typing speed to decision quality, which is exactly where a CTO's cycles belong

🧱 Mission-Critical by Default

Things I have kept alive in production over two decades:

☎️ Telephony switches routing live calls (SS7, SIP trunks, IVRs)
📈 A consumer platform scaling past 4 million users
🏥 Healthcare systems where prescriptions, patient data and pharmacy operations depend on uptime
🧾 Brazilian fiscal infrastructure with zero room for "eventually consistent"
🖥️ 20+ years of Linux, UNIX and BSD boxes, from bare metal to serverless and back to bare metal again

🔬 After Hours

🏠 Homelab: Dell PowerEdge T440, an ARM NAS build on OpenZFS, and 10GbE everywhere it fits
🎛️ Microcontrollers, FPGA, robotics and IoT. If it has pins, I will flash it.
🧪 LLM inference optimization on everything from consumer GPUs to datacenter silicon