Applying Machine Learning to IoT Telemetry
How to handle high-frequency sensor data, filter noise, and build predictive models for real-world hardware systems.
Jun 5, 202410 minMachine Learning
An ordinary Tuesday
Most days start outside, thinking.
Before the laptop opens, there's a walk. Half the model architectures in this archive got sketched out on a phone note somewhere between here and the coffee shop.
In transit
Coffee R&D, officially unofficial.
The elevator ride up is thinking time. It's where the bad ideas get filtered out before they make it to a whiteboard.
Drive path
The commute is a debug loop.
ML.SYS runs in the background the whole way. Somewhere between two traffic lights is usually where the actual fix shows up.
An ordinary Tuesday
Most days start outside, thinking.
Before the laptop opens, there's a walk. Half the model architectures in this archive got sketched out on a phone note somewhere between here and the coffee shop.
In transit
Coffee R&D, officially unofficial.
The elevator ride up is thinking time. It's where the bad ideas get filtered out before they make it to a whiteboard.
Drive path
The commute is a debug loop.
ML.SYS runs in the background the whole way. Somewhere between two traffic lights is usually where the actual fix shows up.
The itch to go further
Somewhere out there, the map runs out.
Turn the page sideways
There was always a bigger map.
Growing up on stories about chasing something impossibly far away leaves a mark. Not the plot. The shape of wanting something that big.
A compass doesn’t care about the coastline.
Most of the interesting work has been in fields with no established path yet — audio ML, embedded vision, wherever the map ran out.
Restlessness is just direction.
The itch to build the thing nobody asked for yet. That restlessness is the actual engine behind most of what’s in this archive.
Also,sometimesit’sjustsilly.
Anime existentialism and nautical ambition don't leave much room for actually being funny. So: a break. Tap one.
The turning pointLeaving the safe path
Leaving the safe path
for the one nobody
had mapped yet.
Trading a settled track for embedded ML, audio systems, and every unglamorous debugging session in between. No guarantee it would work. It was the only decision on this list that actually kept me up at night.
No filter, this time
Same person, no illustration.
The drawings were fun. But under all of them it's just this guy, still figuring it out.
off the clock.
somewhere, mid-run.
no filter, promise.
The captain's log
And every leg of it got written down.
The whole archive, unabridged. Some of it is engineering. Some of it is just thinking out loud. Pull any entry off the shelf.
2 entries
To be continued.
That's the whole story so far — the ordinary mornings, the itch to go further, the jump, and everything written down since. Thanks for scrolling the entire way.