finally! local AI is accessible to everyone.
we replaced our massive smart home server with a single $8 ESP32-S3 chip processing voice commands entirely on-device.
now anyone can build this. here is the full guide:
if breathing gives you away, what about your heartbeat?
a radar placed near the chest picks up the tiny vibrations of each beat. with the right processing, they show up as a signal that follows the rhythm of your ECG
the graph uses real GUARDIAN recordings: orange is the radar, gray is a reference ECG. the 3D scene illustrates the idea
and no, this one doesn’t work from across the room. the radar has to be about 20 cm from your chest
data: GUARDIAN radar vital sign dataset by Kilin Shi, Sven Schellenberger and colleagues, CC BY 4.0
radar: 24 GHz, about 20 cm from the chest. our processing and visualization of one frontal recording, played at 1× speed. the 3D person illustrates the idea
doi.org/10.6084/m9.figshare.…
nature.com/articles/s41597-0…
if breathing gives you away, what about your heartbeat?
a radar placed near the chest picks up the tiny vibrations of each beat. with the right processing, they show up as a signal that follows the rhythm of your ECG
the graph uses real GUARDIAN recordings: orange is the radar, gray is a reference ECG. the 3D scene illustrates the idea
and no, this one doesn’t work from across the room. the radar has to be about 20 cm from your chest
wait. this drone can name every object in your house. in real time
it flies through a kitchen and tags humans, chairs, tables, the oven, each with a confidence score
the model is YOLOv5. "You Only Look Once": it looks at the whole frame in a single pass and returns boxes with a label and a number for how sure it is
that's the purple box around the oven
and it has one bad habit. it keeps calling everything a TV. a detector only knows the classes it was trained on, and TV is a common one, so any bright rectangle in a dark room gets a low-confidence TV label
right now a human still flies it. the AI only watches
the next step is turning those boxes into steering: chair ahead, turn. person ahead, stop
wait. this drone can name every object in your house. in real time
it flies through a kitchen and tags humans, chairs, tables, the oven, each with a confidence score
the model is YOLOv5. "You Only Look Once": it looks at the whole frame in a single pass and returns boxes with a label and a number for how sure it is
that's the purple box around the oven
and it has one bad habit. it keeps calling everything a TV. a detector only knows the classes it was trained on, and TV is a common one, so any bright rectangle in a dark room gets a low-confidence TV label
right now a human still flies it. the AI only watches
the next step is turning those boxes into steering: chair ahead, turn. person ahead, stop
you've seen this bug in every car video
the wheel suddenly spins backwards. the camera takes pictures too slowly for the wheel, so the picture lies
this homemade STM32 oscilloscope has the same limit. it takes 10 million snapshots a second, and past 5 MHz its waves start lying like that wheel
safe up to 3.33 MHz, cleanest at 1-2
every radio signal works by the same rule. my guide ↓
you've seen this bug in every car video
the wheel suddenly spins backwards. the camera takes pictures too slowly for the wheel, so the picture lies
this homemade STM32 oscilloscope has the same limit. it takes 10 million snapshots a second, and past 5 MHz its waves start lying like that wheel
safe up to 3.33 MHz, cleanest at 1-2
every radio signal works by the same rule. my guide ↓
update: my ESP32 classroom now has practice!
a few days ago you could explore the board. now you build the circuit yourself on a virtual breadboard:
> drag a resistor and an LED into place
> draw wires from pin to pin
> press Check circuit, then Run, and watch the LED blink in 3D
> download the .ino once it works
guided mode explains every step, down to why five holes on one row are already connected
3 practice lessons so far: blink, a button and an analog dial. more are on the way
free, in your browser
I just built an ESP32 classroom with Astra
no need to buy a starter kit just to figure out what the pins do, start in your browser:
> explore a 3D ESP32 and hover to learn what each component does
> follow highlighted pins and step-by-step wiring guides
> experiment with LEDs, sensors, motors and a tiny piano
> go from interactive demos to wiring diagrams and Arduino code
16 lessons. Free & open source. Desktop & mobile.
Try it in reply↓
update: my ESP32 classroom now has practice!
a few days ago you could explore the board. now you build the circuit yourself on a virtual breadboard:
> drag a resistor and an LED into place
> draw wires from pin to pin
> press Check circuit, then Run, and watch the LED blink in 3D
> download the .ino once it works
guided mode explains every step, down to why five holes on one row are already connected
3 practice lessons so far: blink, a button and an analog dial. more are on the way
free, in your browser
I just built an ESP32 classroom with Astra
no need to buy a starter kit just to figure out what the pins do, start in your browser:
> explore a 3D ESP32 and hover to learn what each component does
> follow highlighted pins and step-by-step wiring guides
> experiment with LEDs, sensors, motors and a tiny piano
> go from interactive demos to wiring diagrams and Arduino code
16 lessons. Free & open source. Desktop & mobile.
Try it in reply↓
wait, a used phone under $200 is a whole computer?
this is an S22 motherboard with no screen, no battery, just a fan. it runs a desktop on a monitor
same chip class that runs small local AI models. most people throw this hardware away...
wait, a used phone under $200 is a whole computer?
this is an S22 motherboard with no screen, no battery, just a fan. it runs a desktop on a monitor
same chip class that runs small local AI models. most people throw this hardware away...
you can build a dashboard for almost anything
this one is for a rally car: boost, oil temperature, coolant, battery voltage and engine status on a screen the size of a phone case
a small board reads the messages the car already sends around, about 165 of them every second, and draws the numbers live
for everything the car doesn't report, you add your own sensors. my guide to that is below
what would you build one for?
you can build a dashboard for almost anything
this one is for a rally car: boost, oil temperature, coolant, battery voltage and engine status on a screen the size of a phone case
a small board reads the messages the car already sends around, about 165 of them every second, and draws the numbers live
for everything the car doesn't report, you add your own sensors. my guide to that is below
what would you build one for?
this is what a Tesla looks like if you build it from cardboard
no pistons or fuel here, just six tiny electric motors in a V, all pushing one big gear
the V shape is borrowed from V6 engines. a real electric car is simpler: a motor turns the wheels through a gear, with no gearbox full of shifting
the cardboard version still goes brrr
this is what a Tesla looks like if you build it from cardboard
no pistons or fuel here, just six tiny electric motors in a V, all pushing one big gear
the V shape is borrowed from V6 engines. a real electric car is simpler: a motor turns the wheels through a gear, with no gearbox full of shifting
the cardboard version still goes brrr
an ordinary calculator can become a counter for the physical world
it already has a display, arithmetic logic, and a running total
add an infrared sensor, and changes in reflected light can become counting pulses
the trick is the "=" key: on compatible calculators, repeated presses keep adding 1. let the sensor trigger those presses, and the calculator does the counting
the same principle can count passing objects, shaft rotations, or winding turns
that’s a useful starting point for sensor projects:
1 decide what physical change matters
2 turn it into a clear event
3make the device respond
my ESP32 sensor guide below covers choosing sensors, filtering noisy readings, and building that loop yourself ↓
YOUR EYES CAN GIVE AWAY WHERE YOU’VE BEEN
researchers developed an AI system that reconstructs 3D objects behind the camera from reflections in people’s eyes
the surface of your eye acts like a curved mirror.
as you move between photos, it reflects the same scene from different angles
the system accounts for that curvature, separates the reflection from the iris pattern, and combines the views into a rough 3D reconstruction
(the blurry toys in this video are a real reconstruction)
it needs several captured photos with reflections. a recognizable object or interior could then become a clue to where those photos were taken.
YOUR EYES CAN GIVE AWAY WHERE YOU’VE BEEN
researchers developed an AI system that reconstructs 3D objects behind the camera from reflections in people’s eyes
the surface of your eye acts like a curved mirror.
as you move between photos, it reflects the same scene from different angles
the system accounts for that curvature, separates the reflection from the iris pattern, and combines the views into a rough 3D reconstruction
(the blurry toys in this video are a real reconstruction)
it needs several captured photos with reflections. a recognizable object or interior could then become a clue to where those photos were taken.
i hope you will use this for research purposes only...
github.com/HadiZayer/eyenerf




