Skip to main content
Live lab

Field systems and operational telemetry

Live system
Camera relay, telemetry, and edge-AI health in one surface.

Live edge-AI camera and solar telemetry from a board I built end-to-end: hardware integration, Linux services, relay APIs, and the public read-only stream you are seeing.

Hardware integrationEmbedded servicesCamera relayTelemetry UIFailure statesField-camera case study
Edge stack
Linux board, 5 MP camera, on-device inference, thermal/fan health
Power stack
Solar + LiFePO4 telemetry with graceful stale/offline behavior
Web stack
Same-origin Next.js APIs, stream fallback, public read-only surface
AI Readiness
Cam1 model quality is measured with dataset, diversity, and live inference gates.

This makes the edge-AI claim inspectable: the site shows whether labels, class balance, unique views, detections, and RKNN inference are healthy before calling a model trained.

checkingloading gates
Stream
— FPS
RKNN
checking
— ms
Detections
0
last 15 min
Sanitizer
clean-frame age
Training gates
Images0 / 50
Labeled0 / 30
Labels0 / 30
Unique0 / 20
Classes0 / 2
Next best action

Waiting for the camera training service to report its next action.

No collection session reported.

Class coverage
no labeled classes
detections freshservices healthy
Waiting for first sample
camera
Camera loading
Health
Connecting
Thermal
Fan
Camera
RAM
Services
Solar
V
Connecting…
Solar in
W
Load
W
Yield today
Wh
Solar input · 24h
history
Pi 24h history pending
  • 5 MP H.265 sensor, browser-safe preview path
  • Linux board with on-device inference
  • Solar + LiFePO4 buffer, off-grid capable
  • Public read-only stream surface