Interesting. Not clear what it really does. The hardware is an oscilloscope probe on a 3-axis CNC mechanism. That's called a "flying probe", and you can buy one.[1]
Fine. But what does the AI do? It "ingests the project", but what does that mean?
Finding all the pins? That's a start. Using a SPICE model to figure out what should be on each pin, and checking? Now that would be impressive. Probably something in between.
The usual use for this sort of thing is that you probe a known-good board to find out what voltages and signals appear where, and then compare with newly manufactured boards. That's a common production check.
There's potential here. If the AI has some concept of what the board under test is doing, and can diagnose problems, that's quite useful.
I think the novel idea here is you jam some hardware together (whatever you like) that can do "physical real world" things with a well understood interface and then spin up Claude with access to it.
The way I'm thinking about it is, it's a _workflow_ innovation?
So you ask for data sheets for all the visible chips and get PDFs in an output directory with minimal user interaction except to flip the board, ask for a basic idea of connectivity, get a stitched high res surface image etc.... which of course are all currently possible, but you can do them potentially with very low effort. There doesn't have to be a _software stack_ ahead of time. You ask Claude to do the thing, it will figure out how to do it, write some code, pull in some OSS and make the thing happen. You can take this project's software or leave it.
You might say "tell me where you think the JTAG headers are" and it will come up with a workflow to do its best at that task (most likely with variable results...), but nonetheless this is not a thing you can ask of any commercial product I am aware of today. With probes, stuff can get interesting.
Of course experienced hardware & reverse engineers already can do all this stuff and have a plethora of workflows for it but I still think it's an interesting POC of a generalisable approach. You can take or leave this particular software stack. Also, the hardware barely matters, you can duct tape whatever to whatever.
It's lower level than that. "It will probe the approved targets and report back." It has enough smarts to find the pins, and maybe it can read the text labels on some ICs, but that's about it. That eliminates much drudgery, though. And probably the job of some tech who did that by hand.
From my understanding is you’d probe the board during different operations, process the results and deduct what signals are useful and traffic transmitting across the board (I.E private keys, what protocols are used, debug interfaces, firmware components, chip functions, etc).
I believe the standard production check is more like you check continuity between known nets, given that you're the manufacturer, you already know where they are exposed, and therefore you can perform those checks before adding any components. Post component checks are a lot more complicated because active components and passive components will modify the visible voltages and characteristics, often to the point where you won't have the same degree of physical insight.
I would assume once machines are set up that this is only really done if you're not confident of your manufacturing line for some reason (eg. maintenance, reconfiguration) or you are pushing limits somewhere, for example, particularly small vias or traces very close to the edge of the board.
To make this useful, you would want two flying probes because otherwise it's not going to be telling you much you don't already know.
That's a blank board test. This is straightforward given the netlist and Gerber files.[1] It's just a continuity check between known points. Does not require AI.
It’s hard to imagine anyone wanting this to be real more than I do, but this is nowhere close to being ready to do actual work. Photographing real PCBs is hard, there is no fiducial maths, no actual probing is being done. It's just photos of a photo being piped into an agent. If it actually did what it claims, no reasonable person would exclude it from the demo video.
Cool demo, but letting an AI Agent drive a physical probe may have problems. AI is probabilistic, but hardware is precise. If the model miscalculates a pin's position by even 0.1mm, the probe may crush the board. I am curious how the author actually bridges the gap between the Agent’s 'guess' and the sub-millimeter precision needed to avoid damaging hardware?
It's both cool and a bit confusing. Is this an attempt to commoditize flying-probe testing for PCBs? An attempt to use LLMs to reverse-engineer circuits? Both?
It almost feels like it would benefit from being split into two projects. If I'm testing my own PCBs, I probably don't want an agent in charge, at least not routinely. There's just no reason for the added cost, complexity, or non-determinism. And if I'm reversing someone else's design, then going through the effort of building an auto-prober seems like an overkill, especially since a single probe is seldom enough. Even the simplest serial interface will often have one line for clock and another for data, so you're gonna be manually making connections either way.
I figured this was some kind of home-grown prosthetic arm whose wearer could, using AI, draw any artwork online, speak ASL, perform minor surgery on themselves and so much more so long as their supply of tokens lasted.
Perhaps a smidge disappointed when I had a look and discovered it wasn't that :)
Maybe put the probe on a spring loaded linear sensor, and move down until it hits a target offset (could just be read by a simple flag in an optical sensor)-- resulting in hitting a constant target force.
Wow that's nuts. What a great idea! I wonder how much of this the commercial flying probe machines can do already. Pretty cool to be able to have this on a home scale.