In this time weeks feel like months in terms of tech-related news and the new technologies.
I have been fiddling a lot with AI-driven development at work, namely using Cursor and claude all the time. Instead I started code-reviewing even more than ever – not only the code of my human teammates, but not also the code produced by AI.
Everything moves so fast that I just wanted to note a few thing. I’m pretty sure even few months from now it would feel “ugh, so much May 2026”, but still it’s important for me.
In no particular order:
Sourdough IoT sensor
I managed to build and run a sourdough IoT sensor. Okay, this have been in the works for quite a long time (at least since last summer), but I’ve got to admit using LLM sped up iterations unbelievably. Troubleshooting unknown issues now takes minutes – instead of going a long cycle the forum and waiting for a person respond. Bad news for forums, for sure, on the other hand there should be some real communication and sharing happening now instead of newbies asking the same questions over and over again. Was especially surprising and valuable to find it out in the Arduino context.
So what I built? A system is of 4 main components – IoT sensor, IoT message broker, the web-server storing the reading into a database and a mobile application sending push-notifications on important events (like sourdough starting to fall down).
I’m pretty happy with the sensor – based on ESP8266 (Wemos D1 mini) it fulfils the requirements of being ultra-low power (by going into deep sleep between transmissions, which are done using ESP-NOW), uses time-of-light for the measurements, rechargeable, transmits data through the fridge door and can sit on a battery for weeks (estimated).
Message broker is another Wemos D1 (I had lying around) that uses receives the measurements directly with ESP-NOW and then posts them to a remote web-server. Not much to see there, and that was by design.
Then, web-server is just a few PHP endpoints that do some basic checks before storing the data to a MySQL database. The database itself has some triggers to make an approximation across the recent events to understand where the trend goes – that’s enough at this stage to send a correct event, notifying the baker of what’s going on.
All of the above was vibe-coded completely to a great extent – and I can say without shame that for a evening/weekend maker as myself that was an incredible help – both minimising the number of iterations by producing reliable code in the first place, but also coding the whole blocks that I would be bored to death to even start reading the documentation for. Also includes the hardware solution itself.
Now, for mobile application I used Blynk IOT – well, that was too simple to even describe it much. I love their product.
An important milestone for myself – that I’ve made something actually useful and managed to stop at that point not turning it into ever-not-perfect project waiting on a bench. Cool stuff.
Desktop browser game Ansible Deep Field
Gaming has been my thing for a long time. And I always wanted to build a game or two. So this time I was finally ready to pull the trigger and make one.
I call this the game mechanics prototype – I had some ideas in my mind, turned it into playable thingy, tried it with my friends and family, amazing feeling.
8x8x8 LED cube for soldering for reference
The whole idea came from a 8x8x8 LED cube soldering exercise you can order for cheap from China. Then I was also inspired by the Ender’s game book and Das Boot movies for both their cold and silent warfare depictions – both share this deep space/water ambience and lack of visibility for the outer world about the life-changing things happening somewhere.
The game was built using Godot engine and GDscript almost leaving me out in terms of coding since I’m not proficient with Godot at all.
There no graphic assets except for some free fonts and audio – everything is procedural.
I was happy to dip my toes in the game design – as always and what everyone around is saying – it’s still most important to make correct decisions which will shape the final product. And this part is where creativity is at all times high – you can whatever you want and quite fast – so you need to choose wisely, since every decision still needs to be tested and validated.
You can try the game here with a desktop browser.
Actors movie relations
This is just a small thing I always missed on IMDb – going into the movie cast details, and see if any of the cast has acted together before and in which movies. You can often find some of these details in movie trivia, but if none of the nerds was around to post it you’ll never know.
So using claude I built a small web-site (which does not even have a name yet) where you can browser the IMD’s open dataset for movies and then see the connections between the actors across other movies by years. I would still say it’s important those nerds explain the significance of each star pair, but sometimes I just want to see the dynamic between specific actors in another flick.
Fixed my 3D printer issue in klipper
This one was really cool to see working out.
I have an old-ish Ender 3 that I was still using a bit. It had most of the internals replaced already – an SKR Mini V3 motherboard and also a Raspeberry Pi running klipper.
And after replacing the burnt Raspberry for it (or maybe after the experiments with the noise cancellation sensor), the printer’s physical menu button started crashing the firmware. It wasn’t a big deal since I’ve always been using web-interface or sending print jobs from Cura directly via a network plugin, but still unpleasant. Sometimes I just wanted to preheat the hotend when staying next to the machine.
So after sending claude after my menu.cfg files in the klipper directory it managed to fix it from the first attempt. That was a huge win, which otherwise would be quite difficult for me to nail down. Allowed me to close one point with this point – one other home task less.
Misc small repairs and prints
Our oldish dishwasher had the detergent box latch not locking – it simply got it’s latch tooth worn over time. I found a replacement, but it was the bigger module together with another compartment and probably locking electronics and it was selling for 60EUR (which more than we’ve paid for the dishwasher). Of course I said “nah” to that and at some point understood that I should just print the tooth and glue it into the part without replacing the whole latch door. The total printed part is like 6x12mm, and that was a very cost-effective repair from my point. I get really satisfied with things like this. It also doesn’t look ugly, and is a universal remedy to this kind of issues.
Someone had to feed our fish during the vacation. We asked the neighbour, but he had a cat, so that was not really viable. I tried to design my own thing, but I had a very tight time constraint, so I chose this public model at MakerWorld here instead.
I also had a spare micro:bit with a servo-shield lying around, so I ran Claude or cursor over the proposed firmware to convert to javascript and flash to the microcontroller. The thing worked flawlessly – I guess? – the fish survived for more that a week.
