Skip to content
KIMANA.DEV
FR

DomoSo Agri-Watch · Project in development

Designing AI for real-world constraints.

An edge AI project for livestock monitoring, with local processing and deferred synchronisation. My role: founder and CTIO, from architecture to technical decisions.

From field observations to decisions

  1. Livestock observations
  2. Local ARM vision
  3. Local storage
  4. Deferred sync
  5. APIs and dashboards

The architecture separates field processing from cloud consolidation. My decisions cover computer vision, data exchange and continuity when the network is unavailable.

YOLOv4-Tiny · Raspberry Pi / Orange Pi · MQTT · InfluxDB · GCP · Django

The problem and architecture decisions

Intermittent connectivity

Process observations locally and synchronise when connectivity becomes available.

Limited energy

Design for ARM hardware and solar supply; measure consumption under real operating conditions.

Field data

Build a dataset and evaluate computer vision against representative situations.

What still needs validation

Long-term detection performance, energy autonomy, multi-site operation and the usefulness of the data for livestock owners. Potential insurance or financing applications remain hypotheses to validate with the relevant institutions.

Discuss the project

Visit DomoSo