we built a lightning strike detector in just 2 days in partnership with Embedder and open sourced it! the future of hardware is fast. a thread 🧵

1/ it picks up strikes 40km out and puts the distance to the storm front on screen, so you know somethings coming while its still out of sight and earshot. it clips to a pack and runs for months on one charge.

2/ time breakdown - 30 min part selection - 1h30 schematic - 30 min design review - 4h layout (semi-manual) - 4h firmware from @embedder_dev we built a crazy fast pcb design agent that autonomously finds parts, generate schematics, runs SPICE sims and even orders the board. this is what lets us ship boards for our clients so fast!

3/ the AS3935 shaped everything else on the board. a strike radiates a broadband rf burst called a sferic that carries for hundreds of km. the chip listens for it on a tuned loop antenna, measures the shape of the pulse, and estimates the distance to the approaching storm!

4/ the coil is tuned to 500khz, so you have to be very careful to keep your communication protocols and switching regulators out of this frequency range!

5/ we then handed it over to the @embedder_dev agent, which connected to the debugger and started working away. it got the kicad schematics, the project files and a page describing what the device does. in a few short hours, we had fully completed a full hardware project!

6/ we took it 1200m up to wedgemount lake in Vancouver, Canada and took some great photos!

7/ read more about the project or grab the hardware files here! hardware: full writeup:

Thanks for collaborating on this! 🤖⚡️

@tercio_miranda @embedder_dev I built one too…

@tercio_miranda @embedder_dev cool!

@embedder_dev Awesome!

@embedder_dev Add a open server to triangulation from multiple devices and add an api for home assistant.

@embedder_dev don't think you need triangulation actually, the datasheet for the chip walks through how it determines distance. do note that it's not the distance to a single strike, but the distance to the storm (there isnt a strike 1km away 😱)

@embedder_dev also noticed WeatherFlow Tempest is the same chip, so their own distance reports are somewhat based on rough calcs.

@embedder_dev Sickk!!!

@embedder_dev Maybe during the lightning, it could turn off the devices via smart plugs and home assistant to help protect them from the lightning. It could have a timeout period or something to keep the devices off while lightning is ongoing.

@embedder_dev you should build it! everything is open source :))

I call BS. Not because this would be impossible, you can do this in your head even if you count the seconds between flash and thunder. Every 3s is 1km as the speed of sound at sea level is ~333 m/s (light almost instant). I call BS because of the movement of the device (looks like AI, does not move in sync with th hand and because there is no thunder. I don't see how you would accurately determine distance without sound.

@embedder_dev hardware files are on github for you to inspect yourself ;)

@embedder_dev the datasheet of the chip also explains what is being done in detail:

I looked into it tried to understand it and I want to give you the benefit of doubt so I'll say how I think it works and you correct me when I'm wrong okay? I'd imagine it working like this: A lightning strike emmits a broad spectrum of electromagnetic radiation but since it emmits above 500kHz (radio waves), we use this as a trigger for "there was a lightning strike at some unknown distance away". Now we look at the amplitude of that signal which we know decreases over travel distance. Since the amplitude of radiation above 500kHz caused by lightning strikes varies greatly, we observe multiple strikes and try to correct the error that way by making some assumptions about the know distribution of amplitudes in normal storms and the longer we listen, the more accurate we become. If I'm correct in assuming it works like that which might not be the case, I think this would be very error-prone and does not seem to be better than counting time between flash and thunder or do you have data on the accuracy with reliable ground truth? Why does the movement in the video look so unnatural?

@embedder_dev not sure whats going on with the video. if you have doubts you can check the higher quality pics of the device + the hardware files

Is this how it works though? I looked more into it and at 500kHz, the amplitude E decay is proportional to (see image) depending on initial amplitude E_0, distance r, surface conductivity sigma. Sigma varies by a factor up to 20 depending on surface conditions like wet or dry. This means that any distance estimation has an error of up to sqrt(20)>4 and this error can not be corrected at all by listening for more events. Do you have test data with reliable ground truth?

I really don't want to be mean or devalue your work but since it doesn't seem to work, there is a reliable method already without device or cost (counting seconds), I'd recommend you leave it as a lost 2-day project and not put more time into it. All assuming this is really how the method works but it seems like it does.

@embedder_dev Pretty neat stuff. I live in Florida (one of the lightning capitals) and would love to try this out.

@embedder_dev grab the files here!

@embedder_dev Thank you!

@embedder_dev Genuinely impressive turnaround. What's the false positive rate looking like so far? Lightning detection usually lives or dies on rejecting switching noise from nearby industrial equipment curious how you're filtering that out.

@embedder_dev How many tok/s?

@embedder_dev You might want to check which coordinates a network of detectors (which you can join by getting the right hardware) that maps real time lightning worldwide.

@embedder_dev cool!

@embedder_dev it's 1km away?!?! go hide under a rock Daniel 🪨🔥

@embedder_dev yeah we getting fried over here

@embedder_dev Now THIS is what I wanna see

@embedder_dev Very cool!

@embedder_dev Integrate this with home assistant and I’d totally buy one

@embedder_dev project is open source, you should totally do it!

@embedder_dev The only problem with open source hardware, is that you still have to buy the hardware

@embedder_dev detector is nice, but wake me up when you have a predictor.

@embedder_dev @i2cjak

@embedder_dev I love it)

@embedder_dev Def something you can do software-only with the onboard sensors on any phone. It’s novel, but not worth a whole dedicated device. And in the clip, the lightning flashes are CONSTANT and no rain on the window at all. Is any of it real?

@embedder_dev it is real, you can check the hardware files yourself :))

@embedder_dev This is interesting as Fcukkk. Phenomenal job you guys

@embedder_dev In just two days with an IA prompt is very long. Find a faster AI.

@embedder_dev Are there instructions for the hardware?

@embedder_dev yes!

@embedder_dev Very cool!

@embedder_dev Nice. can you put it in a USB dongle?

@embedder_dev I want one.

@embedder_dev I am not living wherever that is.

@embedder_dev 🔥🔥🔥

@embedder_dev Cool 😎

@embedder_dev Hello! Which agent did you use to assist with the development of the schematics and PCB? Could you share it?

@embedder_dev it's an agent we built internally, you can dm me if you wanna try it ;)

The Department head at USF School of Electrical Engineering built a lightening detector. He then spent $45k on packaging and marketing to find out that there is a big difference between Knowing and Caring. He kept one in his office to remind himself that it does not matter how smart you are, if you can not sell the product, it was a waste of time and a waste of money.

@embedder_dev That's just another pointless AI device that will end up in a landfill.

@embedder_dev Cool brah

@embedder_dev Popov would be proud

@embedder_dev what in the world! that is so cool!