A client needed AI-powered incident analysis for their IoT fleet, but security policy prohibited sending device telemetry to any external service. We built it on their existing hardware.
How to design fleet-level health monitoring for thousands of IoT devices — topology choices, data pipelines, and the anomaly detection layer.
How to take a trained model and make it actually run on your constrained embedded hardware — quantization, pruning, and deployment strategies. Includes five failure modes we've seen kill projects after the prototype worked.
Hardware-bound licensing using TPM 2.0 — how to architect it correctly, and the mistakes that cause production incidents.
A practical guide to implementing verified boot on embedded Linux systems — from U-Boot configuration to TPM-anchored trust chains.