Hi!

A lot has happened this quarter; actually, seems like a lot is happening every quarter. Maybe three months is just a long time?

This time I write about moving my studio to the basement, our "smart" home setup, and give you a short experience report from setting up Nix on macOS.

Ever since properly joining the workforce, I've been always working from home, which I acknowledge is a privilege, but not without its downsides. I've heard multiple people struggling with "the couch is right there", but I had the opposite problem: "the work is right there", and there's always something more to do. It was always hard to finish my workday, and the timezone shift to the west coast doesn't really help either (all my coworkers are only just starting their day). Also, home can be distracting sometimes.

For a long while, I dreamed of having my own separate space, one that I can lock up and leave at the end of the day, one that's quiet when I need it, one where I can be noisy when I want to, and one where I can be messy — where I don't have to worry about dropping a screwdriver on the floor, or having to immediately clean up when I drill into an aluminum extrusion.

At the beginning of March, this dream came through. The building we live in had a small doctor's office, which has been unoccupied for a good couple of years now. We managed to rent it, and spent the last year renovating: first demolition and clearing up everything we could, and then light polishing pass — new floor, new bathroom — and finally two weekends of running up and down the stairs to move from here:

To here:

It's only been a month since I've moved, so definitely not enough time for any sort of long-term reflection, but I definitely feel a clearer separation between when I'm "at work" and when I'm not, which is great.

Moving to the basement is a good segue into talking a bit about the minimally-smart home setup we have, as the move allowed me to add some fun automations.

My general philosophy here is that the IoT stuff should have no access to the outside world (internet blocked on the router), and that the automations, if any, should be as straight-forward as possible.

The setup consists of:

• Raspberry Pi (not even one of the new ones) with a Zigbee adapter for one IKEA led strip that we use for ambient light behind our TV.
• DIY mmWave presence sensor wired to Raspberry Pi GPIO, also for our living room. These sensors are better than PIR (though more expensive) since they can reliably tell you if someone is there or not, instead of just letting you know when there's movement detected.
• A bunch of Wi-Fi sockets flashed with Tasmota, and with blocked internet access.
• A couple of Reolink cameras, dynamically blocked from the internet (more on that below).
• DIY BLE beacon detector for the studio made with ESP32 board.
• lgtv2mqtt2 for control of our TV (which also has blocked access to the internet, but is connected to a local network).

Everything is wired with Node-RED, and the state is kept in MQTT, so even if I restart the Node-RED process, the current state stays alive, and I can monitor and access it in multiple ways. We also have a very basic Node-RED Dashboard so we can toggle stuff manually when we're at home (and I can do so while I'm away too, thanks to the magic of Tailscale; I wasn't so cruel as to force my wife to deal with this though).

The only automations we have set up at home are turning the LED on at dusk if anyone is present in the living room, and turning it off at specified time at night.

The studio automations are bit more fun, since I'm the only one having to deal with them if they misbehave:

• I have an ESP32 monitoring for a presence of a specific BLE beacon attached to my keychain, so the system knows if I'm in the studio or not. Based on that, the camera and a dehumidifier turn on or off (in anti-phase to my presence, the dehumidifier is noisy). The ESP32 code is very simple: every 30 seconds it starts a scan looking for that beacon, and once it makes double-sure it's there or not (there are some false negatives), it updates the MQTT state over Wi-Fi, and Node-RED automations take over. Let me know if you have any other ideas for reliable presence tracking. I was thinking about using RFID, but seems too finicky to have to place my keys exactly in the right spot every time, and even more finicky to figure out how to detect them in a long-range (which is possible with UHF RFID, and they can even be positioned in space with additional trickery).
• When I turn on my speakers, my audio interface also turns on — but not the other way around (I might want to work in headphones). This is mainly because the speaker switches are not easy to reach.
• I also have a script on my laptop that starts my music production / music listening software of choice, and makes sure the speakers and audio interface are on.

Going back to our home, the cameras we have in our living room have access to the internet only if we flip a switch in the Node-RED dashboard. We basically "turn them on" only when we know we'll leave our dog inside for a while, or if we're going on holidays. It would be great to automate this, but I can't come up with a way that wouldn't be annoying to my wife, so here we are, flipping virtual switches like cavemen.

This is all done just for additional peace-of-mind, we have a proper alarm that has nothing to do with this system (so I don't break it when I'm messing around with things).

The internet access is blocked on a router, by SSH-ing into it from Node-RED, executing some nvram commands, and restarting the firewall service, it's terrible, but it works (SSH is enabled, of course, for LAN access only). For those of you who are very curious, we have ASUS RT-AX86U (actually two of them, one in our apartment, and one in the studio, hard-wired with an ethernet cable), and the relevant commands are:

nvram set MULTIFILTER_MAC="MAC_1>MAC_2>..." # yes, ">" is the separator
nvram set MULTIFILTER_ENABLE="2>2>..."      # yes, "2" means "blocked"
nvram commit
rc rc_service restart_firewall

I got these keys by running nvram show, changing some options in the UI, running it again, and diffing the output.

The important part is that everything still works even if our internet connection is not working, and even if the local network drops, the only thing we lose is being able to control the IKEA led, everything else is done through Wi-Fi sockets, which have physical buttons on them (though sometimes hard to reach).

Continuing with the nerdiness, working at Replit was a forcing function for taking a closer look at Nix (since the Repls, and our repos, both use it for package management), which, turns out, is actually pretty nice.

I got double-encouraged by Homebrew breaking some Python stuff (again), and one weekend I sat down and migrated to Nix on macOS which was surprisingly painless:

• I started by installing Nix using The Determinate Nix Installer, which has some nice properties over the stock one (like being able to easily uninstall it if I want to).
• Then, I roughly followed this tutorial, stopping at home-manager managing my dotfiles — I might get around to adding this at some point, but I just didn't want to change too much at once.

The "developer experience" of direnv combined with Nix flakes is great: you just cd into the project, and all of the necessary dependencies are not only auto-installed, but also persisted in a fully reproducible way (well, provided you wrote the flake.nix for them, which really doesn't have to be that complicated).

I can now have different pinned Python versions for different projects, which solves my original issue. I also got rid of nvm since I can get specific Node versions from Nix too, and it just feels nice to have all of that taken care of by a single system.

If you're curious about my pretty straight-forward configuration, you can browse through my dotfiles repository.

What I've been reading lately:

• I've been slowly continuing my mathematics thread, first going through half of Mathematics for the Nonmathematician which just wasn't picking up enough speed for what I was looking for, so I switched to Concepts of Modern Mathematics, which has been great so far!
• I also made it a personal mission to finish all the remaining books I had with the phrase "The Mind" in the title: Frames of Mind (a classic on "multiple intelligences"), Models of the Mind (which was a bit too "pop-science-y" for my taste, but could be a good resource to pull some threads from), and Infinity and The Mind, which, as the title suggests, takes a look at the concept of infinity from a ton of different points of view.

On the web:

• School Is Not Enough
• Ink&Switch has been publishing regular, shorter write-ups on their work on Patchwork. For more "productized" approaches to improving version control experience, check out GitButler, and a nice write-up of how one person uses git worktrees. For more "huh that's interesting", check out how Ableton diffs Max/MSP patches.
• More cool concepts for improving the shell experience. On that topic I shipped inline AI generation for shell commands, and have more shell-related things coming relatively soon, like this two-orders-of-magnitude performance improvement.
• Francois has been working on bringing tldraw to VS Code, which looks pretty cool.
• In-depth guide to CNC Lasers.
• Another cute DIY MIDI controller, this one with endless knobs and LED rings.
• There's recently been a bunch of chatting around what a "design engineer" is/does on Twitter (1, 2, etc.), and I somehow even ended up on a list made by Maggie, which felt nice. Thank you, Maggie! I have some thoughts about this, but not that many — on some level, I feel like all the "is X(hamburger / Pluto / ...) a Y(sandwich / planet / ...)?" sorts of questions are exercises in trying to find a single all-encompassing ontology to fit everything, and I'm not alone in believing that there isn't one. Additionally, the only time when defining what I do mattered, was when I was looking for a new job last year, and even then, it was more of a negotiation with whomever was on the other side, trying to build a definition that was useful in that specific context. At Replit (and before that too), I just do what I believe is important, and it doesn't matter much to me if that means writing a proposal, recording a video, designing something in Figma, or implementing it on the "frontend" or "backend" (oftentimes it's all of these).

As always, let me know if you have any thoughts about anything. You can reply to this email directly, or reach out to hi@szymonkaliski.com

And in the meantime, have a great spring, and see you in three months!