Jetson API Field Guide · Source archive

Dancing to the music: Jetson with a Radio (v0.01)

Exactly how far can you push this hardware? Far.

Originally published as a Advisory Hour Substack post on 2026-07-18. This local copy preserves the text and public supporting media alongside the research record.

Original post

Dancing to the music: Jetson with a Radio (v0.01) image from the original Advisory Hour post.
Dancing to the music: Jetson with a Radio (v0.01) image from the original Advisory Hour post.

The more I use a Jetson the more convinced I am that we are severely under-using the capabilities in the bigger name family members. This even includes the 3060-3080 family of GPUs, which I’ve been putting to work on ever increasing and escalating series of AI related tasks that seem to just, well, work.

I learned a couple weeks ago that software defined radio (SDR) is an affordable distraction that I could get into. Not only that, but it’s a distraction I could franken-cobble onto the Jetson hardware. I remember the long drive home from downtown one day, and just was pondering the thought: the jetson has image gen, text to speech, text completion, speech to text, audio processing… what if I put all that AI hardware onto radio signals?

So, I did. The Jetson now has a radio antenna and SDR connection that pipes over an API into a web interface. I didn’t really feel like I was living in the future until looking at the spectral images coming in off the antenna.

Dancing to the music: Jetson with a Radio (v0.01) image from the original Advisory Hour post.
Dancing to the music: Jetson with a Radio (v0.01) image from the original Advisory Hour post.

The Magic of Radio

The coolest experience though is the website. Not only can I tune into different frequencies, or FM stations-but it’s on a website. I’m hosting the website on tailnet. This means that from my phone I now have a portable radio player for any local radio station, or anything else the system might pickup (which I’ll write on this later), and can listen from anywhere. This is pretty great. No subscription required.

The way this works is that the Jetson has a SDR bridge where the frequency data coming in for the SDR is buffered (think like a sprite backbuffer) for 5 seconds, and then this is then streamed up via webrtc over an API that I have setup to a simple website that’s mixed in with my broader “local ai playground” that’s on the jetson. Tailnet functions as a private VPN bridge so I can access this website anywhere in the world, and since I’ve memorized the Starbucks locations in most airport terminals… is a fun little thing.

What got me into SDR was a magazine in Brownsville, Texas. I was down in the area for reasons that’ll take us down a path I don’t have quite the time to get into. However, I did get some great photos. Here’s a photo of a pretty cool place.

Dancing to the music: Jetson with a Radio (v0.01) image from the original Advisory Hour post.
Dancing to the music: Jetson with a Radio (v0.01) image from the original Advisory Hour post.

The SDR story began at the mall. The local mall in Brownsville has a Barnes and Noble. The interior of which is outrageously nice for as small of a town that Brownsville is. In all seriousness there are larger metros with less across the United States. The day had been especially hot, and in need of something between a bar and a pub, I opted for bookseller.

For whatever reason, inspired by the day perhaps, I picked up several magazines on hardware and electronics and just started reading through them. I’m glad I did. One was a UK based electronics magazine, which was especially interesting and I regret that I didn’t have the travel capacity to bring it home with me. The magazine taught me several things, including where to find SDR components on the cheap. I remember sitting back in the chair at that bookstore going, “Wait, hold on. I could do this and just bolt on radio capabilities to the Jetson. It’d take me all of two hours tops.”

Dancing to the music: Jetson with a Radio (v0.01) image from the original Advisory Hour post.
Dancing to the music: Jetson with a Radio (v0.01) image from the original Advisory Hour post.

What AI unlocks at the hardware level is a profound shift in “time to complete a loop”. Before AI, I’d have to invest time into stacks of books, magazines and cross-sectional references with man pages and whatever I could find online. And then building the API and website isn’t trivial either. I’d have enough of a site built that it’d be an interesting project over the span of several weeks. In AI terms, I can connect hardware and get a functioning website the same day generally. Then, each day I wake up I’m gifted with “one new, better idea”. It’s fantastic.

A case in point in this is a caption system that I’m experimenting with. Kyle Sebeysten shared interesting experiments with his audio gear and whisper model work that had me thinking about what automatic whisper captioning and a SDR that crosses multiple spectrums could do. It might make for great random context for my simulated irc channel, so long as they aren’t conjuring ghosts again.

Connected work