I’m analyzing my YouTube watch history in today’s article.
It turns out I watch YouTube for an average of 6.5 hours a day. As for the rating — it’s 0.12, meaning I’ve contributed 12% of the amount of content I’ve consumed back to the community.
P.S. The article includes a link to the repository with the code and instructions — you can gather the same statistics based on your own data.
It turns out I watch YouTube for an average of 6.5 hours a day. As for the rating — it’s 0.12, meaning I’ve contributed 12% of the amount of content I’ve consumed back to the community.
P.S. The article includes a link to the repository with the code and instructions — you can gather the same statistics based on your own data.
Media is too big
VIEW IN TELEGRAM
During my vacation, ants began taking over the area of the gazebo where I’m building a doghouse.
I had to evict them in an original way.
I had to evict them in an original way.
Remember the post about my GPTs for learning English?
I’ve now turned it into a Telegram bot — @positroid_english_bot. I’m positioning it primarily as a helper for Russian-speaking users.
It has the same features as the GPTs:
* You type a word — it gives you a vocabulary card (definition, pronunciation, examples, etc.);
* You type a text in Russian — it translates it;
* You type a text in English — it corrects it.
For now, it's running on the free qwen/qwen3-coder:free (literally the first free model I found on OpenRouter). It has no memory, doesn't know message history — just a bare-bones proxy, so don’t expect too much. It’s just that writing in Telegram is almost always quicker and more convenient for me than loading the bulky ChatGPT interface.
I plan to work more on the functionality, but it’s already decent as is.
P.S. I just found out that my nickname is an actual mathematical term Oo.
I’ve now turned it into a Telegram bot — @positroid_english_bot. I’m positioning it primarily as a helper for Russian-speaking users.
It has the same features as the GPTs:
* You type a word — it gives you a vocabulary card (definition, pronunciation, examples, etc.);
* You type a text in Russian — it translates it;
* You type a text in English — it corrects it.
For now, it's running on the free qwen/qwen3-coder:free (literally the first free model I found on OpenRouter). It has no memory, doesn't know message history — just a bare-bones proxy, so don’t expect too much. It’s just that writing in Telegram is almost always quicker and more convenient for me than loading the bulky ChatGPT interface.
I plan to work more on the functionality, but it’s already decent as is.
P.S. I just found out that my nickname is an actual mathematical term Oo.
Actually, the English bot runs on the basis of my universal AI bot project.
To launch your own bot, just download the binary, specify the Telegram bot API key (from @BotFather), an OpenRouter API key (or any other OpenAI-compatible provider — you can even use a local model), and your own prompt, which will be sent with each message.
By default, it works via Long Polling, so it can run locally — a server is not required.
In other words, anything that fits the task of getting a single-message response from an LLM with your custom prompt can be launched effortlessly.
The repository has details on setup, configuration, and available features.
As time and attention allow, I plan to add support for file exchange, message history, and so on. For now, it only works in single-message mode.
As for development — the project is written in Go. The repository includes full CI/CD setup instructions, how to configure auto-deploy to a server, Docker support, and all that.
To launch your own bot, just download the binary, specify the Telegram bot API key (from @BotFather), an OpenRouter API key (or any other OpenAI-compatible provider — you can even use a local model), and your own prompt, which will be sent with each message.
By default, it works via Long Polling, so it can run locally — a server is not required.
In other words, anything that fits the task of getting a single-message response from an LLM with your custom prompt can be launched effortlessly.
The repository has details on setup, configuration, and available features.
As time and attention allow, I plan to add support for file exchange, message history, and so on. For now, it only works in single-message mode.
As for development — the project is written in Go. The repository includes full CI/CD setup instructions, how to configure auto-deploy to a server, Docker support, and all that.
We've got some kind of gnat invasion in our city—sometimes they even manage to get into the apartment.
I picked up an insect trap for the occasion - it has a UV LED and a couple of high-voltage coils inside. If something conductive flies in between them, it gets instantly zapped.
Well, the LED gave up the ghost, so I had to bring it back to life.
The LED itself was soldered onto the board and was tiny - I didn’t even have a UV one like that lying around. But I did have a 12V UV strip left over from my DIY tanning setup for curing photopolymer resin.
I didn’t have a 5V-to-12V converter either (the trap runs on USB power). I considered desoldering a diode from the strip, but laziness triumphed over doing it the hard way - so I ordered a converter and stuck in a 3-diode piece of the strip.
As a bonus, here’s a demonstration using a macro lens. For the “victim” model, I used a piece of apple peel on a toothpick.
I picked up an insect trap for the occasion - it has a UV LED and a couple of high-voltage coils inside. If something conductive flies in between them, it gets instantly zapped.
Well, the LED gave up the ghost, so I had to bring it back to life.
The LED itself was soldered onto the board and was tiny - I didn’t even have a UV one like that lying around. But I did have a 12V UV strip left over from my DIY tanning setup for curing photopolymer resin.
I didn’t have a 5V-to-12V converter either (the trap runs on USB power). I considered desoldering a diode from the strip, but laziness triumphed over doing it the hard way - so I ordered a converter and stuck in a 3-diode piece of the strip.
As a bonus, here’s a demonstration using a macro lens. For the “victim” model, I used a piece of apple peel on a toothpick.
Today’s blog post is about the universal AI bot project.
But since I’ve already written about it in the channel, I’ll add some extra content. Today I’m heading out to work on my weekend project — not the kind that takes just one weekend, but the kind that takes up all of them. I’m talking, of course, about the composite dog booth. Though realistically, I only manage to work on it a couple of times a month.
I’ve already reached the epoxy stage (because no matter how much paper you glue on, it just doesn’t give you enough rigidity).
I do have some experience with epoxy, but in a different context — I made a countertop for a bar, embedding bottle caps and coffee beans in resin. I wrote about it in the blog.
But the point is, despite the advertised working time, it hardened within 20 minutes — because it overheated.
That’s what happens when you leave epoxy in a big volume - the heat has nowhere to escape, it self-heats and speeds up the reaction. At the peak, it got so hot it even melted the plastic cup.
But since I’ve already written about it in the channel, I’ll add some extra content. Today I’m heading out to work on my weekend project — not the kind that takes just one weekend, but the kind that takes up all of them. I’m talking, of course, about the composite dog booth. Though realistically, I only manage to work on it a couple of times a month.
I’ve already reached the epoxy stage (because no matter how much paper you glue on, it just doesn’t give you enough rigidity).
I do have some experience with epoxy, but in a different context — I made a countertop for a bar, embedding bottle caps and coffee beans in resin. I wrote about it in the blog.
But the point is, despite the advertised working time, it hardened within 20 minutes — because it overheated.
That’s what happens when you leave epoxy in a big volume - the heat has nowhere to escape, it self-heats and speeds up the reaction. At the peak, it got so hot it even melted the plastic cup.
I started watching the series Pantheon — it aired in 2022–2023.
Without giving away any spoilers, the story revolves around an AI that has gone out of control. It's very much in tune with the current times. Moreover, all the technical details are portrayed quite convincingly for a series of this kind.
But that’s not the point — in the very first episode, they showed one of the characters’ desks, and on it was the exact same soldering iron I’ve been using for about 10 years now — some Chinese no-name model with an analog temperature dial and a long heat-up time. I occasionally look at modern replacements, but this one still works just fine.
Without giving away any spoilers, the story revolves around an AI that has gone out of control. It's very much in tune with the current times. Moreover, all the technical details are portrayed quite convincingly for a series of this kind.
But that’s not the point — in the very first episode, they showed one of the characters’ desks, and on it was the exact same soldering iron I’ve been using for about 10 years now — some Chinese no-name model with an analog temperature dial and a long heat-up time. I occasionally look at modern replacements, but this one still works just fine.
Media is too big
VIEW IN TELEGRAM
I’ve started working on a new project (as if I didn’t have enough old ones already, right?).
It’s tied to Halloween — not sure yet what the final result will be, but something is already starting to take shape.
It’s tied to Halloween — not sure yet what the final result will be, but something is already starting to take shape.
Yesterday on the screen — today in hand. Version two, optimized.
The first version was designed for ES9251 II servos — the smallest ones I could find. Also, the previous model didn’t have a well-thought-out mounting system for the eye.
The current version is made for the most common servos — SG-90. Thanks to a different layout, I managed to keep the same eye size — still 42 mm in outer diameter.
The longest and most challenging part was figuring out how to mount the servo in a way that’s both easy to print and secure. I came up with two additional mounts that slot into the eye housing using dovetail joints. This also makes replacing the outer shell quick and convenient.
Tomorrow, I’ll start working on the software.
The first version was designed for ES9251 II servos — the smallest ones I could find. Also, the previous model didn’t have a well-thought-out mounting system for the eye.
The current version is made for the most common servos — SG-90. Thanks to a different layout, I managed to keep the same eye size — still 42 mm in outer diameter.
The longest and most challenging part was figuring out how to mount the servo in a way that’s both easy to print and secure. I came up with two additional mounts that slot into the eye housing using dovetail joints. This also makes replacing the outer shell quick and convenient.
Tomorrow, I’ll start working on the software.