Software engineer working where software meets machines. I build systems for CNC machine tools and the shop floor around them — serial protocols, the tablets and desktops that drive them, and AI on top of technical archives. C++, Python, and the parts of a stack that break when real hardware is on the other end.
Based in Brazil · open to remote roles worldwide
These three projects are one system, taken apart. A CNC machine needs programs sent to it, needs them sent from somewhere, and needs fixing when it stops. I built all three sides at SMI, where they are in daily use — real work on real machines, not yet sold as products.
| dnc-serial-engine build |
C++17 Qt |
RS-232 DNC engine. Adaptive flow control that streams G-code to controllers with kilobytes of buffer and no telemetry — fast enough to be useful, slow enough not to drop a block mid-cut. |
| dnc-android build |
Qt 6 QML C++17 JNI |
The same engine on a tablet bolted to the machine, reaching RS-232 through USB OTG and an FTDI chip. Ported unchanged behind an extracted ISerialPort — then made to actually send, through five faults that raised no error. |
| maia-cnc-assistant tests |
Python FastAPI RAG Flutter |
RAG assistant over a company's technical archive. Multi-stage answer pipeline with safety policy, human-reviewed ingestion, three-level AI cost guards, Flutter app and WhatsApp channel. |
DNC operation screen with the RS-232 line signals monitored live MAIA knowledge base: 18,299 documents processed and 1,038,300 passages indexed
Left: the DNC feeding a machine, with the eight RS-232 signals that tell it whether to slow down.
Right: the archive behind the AI assistant — 18,299 documents, 1,038,300 indexed passages, 14 machine brands.
If you only read one thing, read the engineering log in the first repo — the flow-control debugging history on live machines, including the fix that caused the next bug.
Systems & embedded — C++17, Qt/QML, Android (JNI, USB host), RS-232/serial protocols, FreeRTOS, CMake Backend & AI — Python, FastAPI, PostgreSQL/pgvector, RAG, OpenAI & Gemini, Docker Apps — Flutter, Qt Quick, TypeScript Ops — Caddy, CI, versioned migrations, backup and restore
I ship things that run unattended in places where failure is expensive, and I document what I got wrong on the way there. The interesting part of a system is usually the constraint nobody wrote down — the controller that lies about its buffer state, the archive nobody can search, the budget that a retry loop can burn through overnight.
Open to remote engineering roles worldwide — backend, embedded and industrial systems, or applied AI. Available for contract or full-time, and comfortable working across European and American time zones from Brazil.
📫 kaikiiquadros091o@gmail.com · 🌐 dev-kaiki.github.io · 💼 LinkedIn