Inspiration
When the original controller of my solar thermal system failed, I decided to build my own replacement. I am a structural engineer, not a professional software developer, but I had a clear concept, an engineering plan, and OpenAI Codex.
My goal is a secure, local-first home automation ecosystem that remains fully operational without cloud services or an internet connection.
What it does
The system autonomously controls and monitors HVAC equipment, including solar thermal collectors and a heat recovery ventilation unit.
A custom ESP32 service dongle collects RS485 and ESP-NOW telemetry, machine registers, system status, and alarm history. Codex with GPT-5.6 helps interpret this raw data and explains faults in natural language, acting as a virtual service engineer.
The demonstration shows a real RS485 communication fault, its detection by the controller, analysis of the diagnostic data, and restoration of normal operation.
How I built it
The system uses ESP32 microcontrollers programmed in C++ with Arduino. Devices communicate locally through RS485 and ESP-NOW. The touchscreen controller provides live status, configuration, alarms, sensor readings, graphs, and autonomous control.
I designed the hardware architecture and operating logic. Codex supported implementation, debugging, protocol analysis, testing, and development of the service dongle.
Challenges
The main challenge was integrating several communication protocols, sensors, actuators, safety mechanisms, and noisy real-world hardware. Diagnosing intermittent RS485 faults from raw registers was especially time-consuming.
Another challenge was developing a large HVAC controller as an independent builder without a professional software background.
What I learned
AI is most useful here as an engineering collaborator, not as a replacement for local automation. The controller remains deterministic, autonomous, and cloud-independent. Codex accelerates development and makes complex diagnostic information understandable.
What's next
The next stage is further development of the recuperator controller, which combines multiple modules and more than 20 sensors. I also plan optional Home Assistant integration while keeping all essential control local and independent from the internet.
Log in or sign up for Devpost to join the conversation.