ls -la ./blog/
Blog
Notes on IoT fleets, telemetry pipelines, cloud infrastructure and applied AI, written from the engineering side by Bartosz Szczepanik.
# 8 posts · newest first
What CRA compliance actually takes for a small hardware team
I turned every Cyber Resilience Act obligation for device makers into one checklist. What it asks of a connected product, and what I would do first.
One golden certificate for the whole production line
Give every device its own X.509 identity in AWS IoT Core without flashing certificates unit by unit, using a shared claim certificate and fleet provisioning.
The CRA clock is already running on devices you shipped years ago
Since 11 September 2026 the CRA's 24-hour reporting duty covers every connected product on the EU market, old ones too. What it means for a small hardware team.
Updating firmware on fifty thousand ESP32s without bricking any
Over-the-air updates for a large ESP32 fleet: partition layout and rollback, signed images, staged rollouts and the cost of sending a megabyte to every device.
Where the battery goes on an LTE-M sensor
The power budget of a battery powered LTE-M sensor, why the radio dominates it, and the settings that decide whether a battery lasts one year or five.
So everyone is building LLM trading bots now
I kept hearing about language models that trade on their own, so I dug into how these bots work, where they fall apart and what I would build instead.
JEV: a model that only decides
TypeSafe AI's JEV returns typed answers with calibrated probabilities and never writes a word. What it is, where it fits, and what to be careful about.
Storing a reading a minute from 100,000 devices
What it actually takes to ingest, deduplicate and keep a year of per-minute telemetry from a fleet of a hundred thousand devices, and where the costs hide.
tail -f ./blog — more posts soon