Live hardware system
Edge-AI Field Camera
Solar power, embedded inference, resilient camera transport, and protected operational proof.
One system. From the physical world to the browser.
Edge-camera system
On-device RKNN
Solar telemetry
Protected playback
01 / Scope
What I owned
Hardware integration, embedded services, relay architecture, telemetry UI, security boundary, and operations.
Problem
Expose a real solar-powered edge system publicly without leaking private infrastructure or letting live hardware failures become invisible.
Built
Integrated camera streaming, on-device RKNN inference, solar telemetry, thermal and fan health, Cloudflare relay routing, and production diagnostics.
Outcome
A live hardware system that behaves like a maintained product: opt-in streams, health fallbacks, quality checks, recovery paths, and deploy-time regression gates.
Engineering constraints
The conditions that
shaped the system.
The field node must tolerate intermittent power, address changes, and upstream service restarts.
Public visitors need useful proof without receiving camera credentials or infrastructure tokens.
Media transport must degrade through WebRTC, HLS, MJPEG, and snapshot fallbacks without hiding failure state.
02 / Key decisions
Tradeoffs made explicit
Private media edge
Camera media and physical writes require signed access sessions; server-held relay credentials never reach the browser.
Always-on relay ownership
The camera path terminates on a dedicated relay host instead of depending on a personal laptop.
Observable failure modes
The interface distinguishes stale telemetry, media failure, service failure, and training readiness instead of showing one generic offline state.
03 / Under the surface
Architecture
- Linux camera and inference node
- go2rtc media and recovery services
- Authenticated Cloudflare media boundary
- Next.js diagnostics and telemetry UI
Validation
- FPS and bitrate budgets
- RKNN state and latency checks
- Solar freshness monitoring
- No-idle-stream policy
- Embedded Linux
- RKNN
- WebRTC
- Cloudflare
- Tailscale
- Next.js
- Victron
04 / Result
Demonstrates embedded Linux services, power telemetry, camera transport, edge inference, and full-stack operational visibility in one public system.
Open the live labEVIDENCE & BOUNDARIES
What this work demonstrates.
The public lab separates service health, solar readings, inference state, and protected camera playback. The system guide remains available without credentials or a connection to the field hardware.
Stream by default, or make media explicitly opt-in?
- Decision
- Keep the public explanation available to everyone, with camera media behind the existing access boundary.
- Tradeoff
- Visitors cannot immediately see private footage. The guide and separate telemetry make the architecture understandable without opening that boundary.
Scope & limitations
A reachable website does not prove the camera, inference engine, or power source is healthy. Use the dashboard’s individual states and reading ages. No uptime percentage or field reliability benchmark is claimed.
- Explore the public lab
Read the guide, inspect current states, and distinguish fresh readings from stale or unavailable sources.