Embedded · IoT · Machine learning · End to end
IoT energy and machine-health platform, from circuit board to cloud
Sensor hardware, firmware, gateways, automated control and a machine-learning analytics platform that learns how rooms and machines behave, cuts energy use and predicts maintenance needs.
- 6
- analytics layers, from raw telemetry to maintenance recommendations
- ≥ 0.90
- macro F1 for the fault-severity classifier (gradient boosting, cross-validated)
- PCB → cloud
- hardware, firmware, gateway, control, analytics and apps
- 8 mo
- of power, temperature and humidity data from live sites
01
The problem
In commercial buildings, cooling and other machines dominate the energy bill, yet most run uncontrolled and unmeasured and are serviced only after they break down.
02
What I built
- Automated control algorithms that run machines against learned room behaviour to cut energy without losing comfort
- A six-layer analytics platform: telemetry quality gates, room thermal models, baseline curves, anomaly detection, fault-severity classification and maintenance insights
- Machine learning: gradient boosting (LightGBM, exported to ONNX) for fault severity, DAMP and ECOD anomaly detection, and adaptive conformal calibration to keep false alarms at a set rate
- Sensor-tag hardware (KiCad board, two revisions, FreeCAD enclosure) with control and vibration-sensing firmware
- Gateway bridging BLE sensors to the cloud over MQTT with TLS; cloud dashboards, alerts and a mobile app
03
Engineering
- Control algorithms and the analytics platform designed and built solo; leading a team of seven across the other layers
- Daily model retraining with automatic quality gates and rollback; models served from ONNX
- AWS infrastructure as code (ECS, load balancer, CloudFront, WAF, managed database) and eight CI/CD workflows