We do not over-engineer. We choose technologies that work reliably in remote field environments, can be maintained by non-specialist teams, and improve over time as data accumulates.
Physical components designed for durability, low power, and remote deployment.
Langstroth-compatible hive body. Hardwood construction. Sensor mounting point built into the base board. Designed for standardized stacking and transport.
Low-power internal sensor. Measures temperature (±0.1°C), relative humidity (±2%), tamper (reed switch), and tilt (accelerometer). Solar-charged LiPo cell. Reports every 15 minutes.
LoRa receiver aggregating up to 100 sensor nodes. Cellular primary backhaul (4G). Starlink failover port. Solar-powered. Designed for outdoor mounting on site perimeter.
The platform stack, built for operations at scale.
Gateway data arrives via cellular/satellite. Parsed, validated, and stored in time-series format. Anomaly rules applied in real-time. Alerts routed to the operations platform.
Web-based multi-role platform. Operator and admin views. Mobile-responsive for field use. Works on 3G connections. No specialist hardware required.
→ Try the demoFull chain-of-custody from site to product lot. QR code generation for end-to-end traceability. Export-grade documentation built from operating data.
→ View traceabilityAI that is honest about what it can and cannot do at this stage of the dataset.
| Capability | Phase | Status | Notes |
|---|---|---|---|
| Threshold-based anomaly detection | Phase 1 | ✅ Active | Rules-based, tunable thresholds |
| Device silence detection | Phase 1 | ✅ Active | Configurable silence window |
| Tamper event grouping | Phase 1 | ✅ Active | Time-window based grouping |
| Multi-sensor anomaly scoring | Phase 2 | ◷ In development | Requires larger dataset |
| Predictive maintenance hints | Phase 2 | ◷ In development | Pattern recognition in progress |
| Colony behavior models | Phase 3 | ○ Planned | Requires validated dataset |
| Disease risk scoring | Phase 3 | ○ Planned | Academic collaboration required |
We do not overclaim AI capability. Phase 1 is rules-based and works. Phase 2 and 3 require more data and more validation. We will be specific about what each model can and cannot do.