Antoine Sauvage · Tokyo
I started with building automation and SCADA, then moved through IoT and edge computing to video AI on NVIDIA Jetson. Konekuto is where I share the independent tools and small products I build alongside my professional work.
What I work on
I lead products and engineering teams, but I also continue to work directly on architecture, code and debugging. On edge systems these parts are closely connected, so I prefer to stay involved from the product decisions down to the implementation.
Independent projects
Some start from a practical problem I need to solve; others are a way to test an idea with a working implementation.
Open source
I built this because testing video AI with physical cameras is not always practical. It turns recorded videos into repeatable RTSP streams and can simulate common problems such as noise, frozen video, interruptions or unstable connections.
View the source →Product
A small web application for finding and checking DENSO Mapcodes in Japan. You can search for a place, select a position on the map and share the result from a phone. It is built as a Vue PWA with Leaflet, geocoding and a Cloudflare proxy for the external lookup.
Open Mapcode Search →Engineering lab
An experimental environment where I test how local AI agents can work with edge and IoT services. It brings together Node-RED, RabbitMQ, monitoring, a small management API and secure remote access, with Compose stacks for ARM64 and x86.
Professional work
This work belongs to the companies and teams involved. I include it here to give context to my engineering background and current work.
EDGEMATRIX
Product and engineering leadership
Edge AI Station is a platform for running video AI applications on NVIDIA Jetson devices near the cameras. I work across product and engineering, including video pipelines, application packaging, device management, monitoring, remote operation and the cloud services used to manage deployed systems.
This work follows earlier projects I did around containers and edge computing. The main challenge is not simply running an AI model on an edge device. It is keeping cameras, applications, updates and remote support reliable once the devices are deployed in the field.
Public reference →Veolia · Sanko · EDGEMATRIX
Engineering, architecture and product work
Before video AI, I worked for many years with building automation and energy systems: BEMS/BACS, SCADA, controllers and the protocols used to connect equipment with supervisory systems. This experience is still useful today because an edge AI system rarely works alone. It usually has to connect with the existing operational environment.
One public example is the integration between Edge AI Station and Shimizu’s DX-Core building platform. Node-RED receives AI inference events and publishes them through MQTT, allowing results from video AI to be used by the building system.
Public reference →Writing
I wrote these in 2019 while experimenting with Kubernetes, microservices and GPU workloads on small edge devices.
About
I started my career in building automation and energy management. The work involved controllers, field protocols, supervisory systems and equipment expected to remain in operation for many years. I later moved into IoT platforms, cloud services and containerized edge computing.
Video AI brought cameras, GPUs and real-time pipelines into the same type of environment. Today I lead engineering around Edge AI Station at EDGEMATRIX, while remaining involved in architecture, implementation and technical problem solving.
Konekuto is separate from my employer work. I use it for open-source tools, small applications and technical experiments that I can develop and publish independently.
Contact
If you want to discuss one of these projects, open-source work or a relevant professional opportunity, you can contact me here.
Contact me