i’ve gone completely fucking insane from music.

18 machines between september 4 and september 30. 610 commits that month, 136 of them on the 19th alone. a quarter of my messages went out between two and seven in the morning. at one point i was telling a kick generator that we would save everyone whose kicks got hurt.

so i’m taking a few weeks off. no tracks, no new machines.

the idea behind all the schizo stuff: take one kick and make every instrument of the track out of it, with almost no processing. you cut tiny pieces of the wave. stretch them, pitch them up, stretch again, chop, consolidate, bounce, chop again, until the waveform turns into something strange.

in ableton that’s slow. every round is one more bounce. so that night i asked for a tool where the waveform is the interface. you put a point on the wave and pull. fewer numbers, fewer knobs. layers you can see stacked, what sits over what. a sample you can stretch forever without baking it again each time. the original is never touched: the tool keeps your moves as a list and renders them from the source every time.

the ask was far from modest. the punchiest kicks on the planet, proven with math. one hit that already carries its body and its bass, so the monitors shake the room and it feels like that came from the strike itself.

so PUNCH·77 has a ban: no sub layered underneath. one wave. you pick a note and a search goes digging. it throws kicks and keeps a front of the best ones for punch, depth and loudness, and every candidate is ranked against a pack of 1092 real kicks.

i watched Anthony Marinelli program a kick on an ARP 2600. his point: if you can send an envelope to the pitch of an oscillator you get the swoop every kick has, and from there you can make any kick you can imagine.

first i turned the video into a lesson for myself. that afternoon i asked for the next thing: cross that patch with MATH·77 and build an instrument for analog-style kick monsters, where every kick already hits the moment it comes out. plan first, then build.

i had one of my own kicks measured: how much bass, how hard it hits. then i asked if code could make it wider and heavier without wrecking it.

what came out of that question was a harmonics generator. a waveform written as a list of partials. you build the tone that sits under the kick one partial at a time, and you hear something i’d never seen so plainly: phase eats headroom. same partials at the same levels, a different peak.

i asked where one of the starting sounds came from, and the answer was: it’s a formula. no sample, no patch. a function of time.

i wrote back in capitals that this is what i need. make a generator of kicks like that, as its own instrument.

MATH·77 is a kick written down. the whole hit is one closed formula of t. there’s nothing random inside, so the same line gives the same kick forever, dry and exact. it’s tied to the bpm. a fold control takes it from a clean note to something close to noise, still without a single random number.

morning, no sleep, and i typed the angriest question of the month. why do i have a great kick generator and no generator for dirty industrial synths or hard broken percussion?

LAB·77 exists to test one hypothesis: can these machines make sounds nobody else has, with the distortion and the dirt already in them, for more than kicks. i asked for no design at all. a couple of parameters at most, and pure random. i only wanted to know if it’s possible.

i used to play live with ableton and an akai. that afternoon i started wondering what my own program in the browser could give me. i already had a pile of machines that make the sounds. maybe they could also save me the time i spend preparing samples.

decided in the first hour: laptop only. no midi to begin with, just keys. the tools on the shelf don’t fit as they are, they’re the base for live versions. hard industrial techno, 150 bpm.

the pattern generator inside BOCHKA RAPIST+ had a flaw i couldn’t live with. the fills squeezed the sound. the kick in the pattern was no longer the kick i’d made.

so PAT·77 has one job, and the job is a refusal. drop in a finished kick. it goes on a grid with rolls and ghosts, and nothing gets processed. every piece is cut from the same hit. i asked for it plain, the basic look, no decoration.

late that night i asked for something fast. a melody generator built from odd algorithms, notes that come out of delays, cuts and reverb. that became GROVE·77, a melody garden: plant a seed and it grows a line with echoes that climb.

i called its output garbage the same night. it’s still on the shelf.

at five in the morning the better idea showed up, and i had a working name for it right away: PERK-ATMO-NOISE. you give it a kick or any sound, and a bpm. it works out roughly what key the sound is in and grows a groove around it, mostly out of noise: percussion, rolls, an atmosphere that can be long or broken or stuttering, noise one-shots thrown into random places. the only law is 4/4. what comes out is stems.

after the codec thing i said what i wanted in one line. a very strange kick. a data kick, a schizo kick. but the punch and the power have to be there.

the first version knocked my speakers onto the floor. so the first feature request was a limiter.

then the real problem: how do you make a kick weird without turning it into dumb noise? the data band just screamed. every kick felt like it grew from the same base. so it got more bases, a MUTANT one that isn’t predictable at all, new distortions (one is called SCHIZIFY), particles flying around inside the rumble, heavy low metal. i took apart three kicks i love from other people’s tracks and used them as references.

i asked for a speech generator. what i wanted was chatter: phrases coming from everywhere, left, right, and one pressing on you from the front.

it worked within the hour. then it got its name, VOICES IN MY HEAD, and my face in the middle of the radar. the more voices talk at once, the more the face tears, and the more of a tribal sigil grows around it. the sound has INSANITY knobs: SHRED, DEMON, RING, METAL THROAT.

i asked what small tools were still missing. two rules for the list. what usually costs money, i make free. and i don’t care about old browsers, i want the newest browser apis pushed as far as they go.

first pick was STEM-77. drop a track, get vocals, drums, bass and the rest, all computed on your own machine. my first vocal split had the instrumental bleeding through. the cause was dumb: a window size copied from someone else’s config, 7680 where the real model wants 6144.

my PX exports kept coming out soft. the 3d renders and the text suffered most. so i wanted an upscaler in the same spirit as everything else here: no server, nothing uploaded, the picture never leaves the browser.

three engines. RESAMPLE is the plain canvas stretch. SHADER is the classic math on the gpu (lanczos, mitchell, xbr for pixel art) and it’s fast enough to play video live. NEURAL runs real networks inside the tab. only the smallest one, 4.6 mb, loads by itself, so nobody sits waiting for a download they didn’t ask for.

the september 26 ritual event got canceled. timing didn’t work out, plus prizrak couldn’t get time off work.

probably for the better. we’ll do it right when the moment’s actually there.

i remembered i’d already built this once. a webcam destroyer, a trippy effect for performing, sitting in another project’s folder. i dug it out and put it on the site.

first real test: i switched it to the microphone, the room was silent, and the picture was still shaking. that’s a contradiction. zero sound means zero movement. that rule ate the whole evening.

what i came up with is a baseline. press a button, it listens for a few seconds and remembers how loud the room gets. everything after that is measured against it. at 90 to 100 it goes off. at 0 to 10 almost nothing moves.

been building the whole visual package for our understav crew’s ritual event. logo, 3d poster, 2d poster, dj posts, teaser, all of it.

here’s the 2d poster. industrial goth meets techno ritual aesthetics. layered cyrillic type, distorted geometry, red-black palette with this cold blue accent in the center. kept it dark but structured, like the lineup itself.

the logo and 3d stuff push it further into bio-digital territory. building a coherent system that works across formats: instagram posts need to read instantly, a4 posters need to pull you in from across the bar.

i keep a private vault. everything lands there through a telegram bot: voice notes, links, half-formed thoughts, whatever i mumble into the phone while walking. that part works fine. this site sits still for a month at a time, because moving something from the vault to here means rewriting it, tagging it, committing it. even asking an ai to do it means opening a thing and explaining myself, and i never feel like it.

i orchestrate a pile of agents on big projects now, rave land and the school platform mostly, and i keep hitting the same wall. i see code appear. i don’t see the path. what they researched, what they tried and threw out, where they’re stuck, what they decided and why. all of it exists, in fifty markdown files nobody reads twice.

what i want is one screen over that. every component with a status: researching, building, works, blocked. blockers with numbers on them so i can point at one in a message. the last ten findings, not the whole journal since day one. goals, with what’s actually done. open questions sitting there waiting on my call. and a decisions archive, so in a month i still know why we took three.js and not babylon.

the pipeline i want for props: point a phone at a real object, and a few minutes later it’s in the library, ready to drop into the world. no downloading, no importing, no fixing normals at 2am.

scanning is the solved part. lidar on a pro iphone through polycam or scaniverse, photogrammetry through luma or meshroom if there’s no lidar. out comes a glb with a texture. for a stylized world it doesn’t even have to be clean, a rough scan after retopo does the job.

been thinking about documentation wrong this whole time. i treated it as a description of the code. when an agent does the work, it should be a panel where i watch the thing think.

code is the only output i get. the process evaporates. which libraries it rejected and why, where it hit an incompatibility, what it learned reading someone else’s repo, which parts are still half done. next session it doesn’t remember any of that either, so it re-researches what it already knew last tuesday.

been pulling animations out of gta san andreas for my own characters, and the format taught me something i keep repeating to myself since.

quaternion, time and root movement sit in the file as integers, divided by constants. every write-up out there says the time divisor is 1024 or 2048. take those and the whole file, fifteen hundred animations, lasts 0.36 seconds. absurd, but the absurdity only shows up if you stop and ask what the number means.

spent an evening digging through vk threads for stock gta san andreas files. this is the actual state of the art for 20 year old game assets. some guy’s forum topic, a rar, no readme, and the page only opens if you’re logged in.

got the standard skins out of it. 593 files, dff and txd pairs, one of each per pedestrian, 5.9 mb for the models folder. geometry and textures, all renderware.

while building the rave world i’ve burned time on: how to generate a world, where to get models, which generators are worth anything, how to rig them, which library normalizes what, which tools animate locally. all of it inside one repo. none of it written down anywhere except my head.

a normal repo keeps code, tests, a readme describing the finished thing. the path disappears. in three months i won’t remember why i took that library over the other one, and i’ll google the same comparison again. anyone else walking in has no chance at all.

found a model aggregator worth writing down: grsai. one key, and behind it both image models (gpt-image-2, the whole nano-banana family, 1k/2k/4k) and llms (gemini 2.5 and 3.x, gpt-5.4, gpt-5.5). the dashboard counts credits and bills in rubles, which is the actual reason it’s interesting to me.

images run 600 to 18000 credits a shot, somewhere between 2 kopecks and 1.8 rubles. text is 0.25 to 27 rubles per million tokens depending on the model. failed generations give the credits back.

idea i want to build a track around. a vocal line runs normally, and then one tiny piece of it, a single letter, a slur between two vowels, gets grabbed and looped. that loop turns into a synth. pad, lead, whatever it wants to be. it lives for a few bars, develops, then the loop lets go and the voice finishes the sentence like nothing happened.

you don’t hear two layers, voice plus synth. you hear one material falling apart into molecules and assembling back.

small arrangement trick i keep using. instead of keeping fills on their own track, put the fill at the very beginning of the sample file, half a second of it, then the main loop right after. when you want the fill, retrigger the clip. playback jumps back to zero, you get the fill, the loop carries on.

one file instead of two, one track instead of two, and every variation of that sound lives in the same place. in a live set it’s a restart on a midi button, no track switching, no scene juggling.

thought that won’t leave me alone. a page can carry two layers of text now, one for people and one for crawlers, and an ai reading that page doesn’t tell display: none from a headline. it takes the dom, comments and hidden blocks included.

old seo spam did this to fool search engines and google eventually learned to punish it. agents haven’t learned anything yet. a hidden line dressed up as normal page content can just redirect the thing reading it.

read a piece on schools and attention that reframed the whole thing for me. the line that stuck: don’t say we teach the children of the future, say we teach children to keep thinking for themselves in a world where thousands of systems try to think for them every day.

that’s a mission you can measure, and it doesn’t expire when the technology changes. it also fits an ai-personalized school without the usual trap, because the tech is there to build independence instead of maximizing engagement.

docs framework worth a look: blume. drop md and mdx files in a folder, run blume dev, and you have a documentation site with navigation, search, dark theme, seo and thirty-something components. no starter to clone, no config to write.

the ai side is why i actually tried it. it generates llms.txt, ships an mcp server so an agent reads the docs directly, and has an ask ai button built in. hayden bleasel from openai shipped 1.0 in july, mit, astro and vite underneath, node 22.12+, blume eject when you outgrow it. content can come from the filesystem, github, sanity or notion, and it renders openapi references and an auto changelog.

voice typing note. people kept telling me to put gigaam v3 from sber into handy, it benchmarks best for russian. tried it. parakeet tdt 0.6b v3 is still noticeably better for me, especially when english words show up inside russian sentences, which for me is every second sentence.

the example that settled it: gigaam hears python and types piton. parakeet types python.

parakeet is 705 mb against gigaam’s 151 and i don’t care. when the text comes out right the first time, voice input beats typing. when i have to clean up after it, the whole advantage is gone.

got a piece of feedback at work that i keep chewing on. i’d fixed some broken links on a site by hand. the reply was that the job wasn’t the links, it was a check at the hugo level that catches every dead link in the build.

so the fix is a symptom and the guard is the cure. a rule without a guard is a wish. if you want it held, put it in the harness: a linter in ci, a commit-msg hook, a build that fails loud instead of a page that quietly goes empty.

PX-77 showed up on july 16, the same day as DOT·FIELD and the tools shelf itself. first version: feed it a photo and the data rots. broken jpeg, sorted pixels, a buffer knocked out of line. a day later it could mosh, mask, bake and film from the phone camera.

then i used it for real and got bored. the degeneration was predictable. same photo, same slider, same kind of damage.

the first VC-77 was a vocal fabricator. you typed a phrase and it printed 16 takes locked to a grid, with a bpm, a seed and a destruction amount on every card.

i hated using it. nothing made a sound until it finished rendering, and i was reading numbers when i wanted to hear a voice.

two days later i threw the grid out and rebuilt it as a voice box. type, it talks right away. loop it. twist the effects while it runs. save the wav when it sounds right. the wav goes through the same chain you were just twisting, so what you heard is what lands in the daw.

the glitch engine had a bug i kept ignoring. paste one of the heavier presets into telegram or instagram and half the marks turned into empty boxes. not glitchy, just broken.

turns out it was pulling combining marks from all over unicode. enclosing circles, the symbol-overlay block, tibetan, hebrew, arabic, even the cyrillic millions sign. looks great in a desktop browser that has every font installed. but a phone only has the fonts it ships with, and telegram and instagram render with those. no font, no glyph, just a tofu box. instagram also quietly deletes marks it doesn’t feel like keeping.

it’s happening. august 7th i’m finally playing at monasterio. excited, a little scared, and ready to show what i’ve actually learned.

it’s been around two, three years now since i started with music, producing, djing. felt like a long build up to a moment like this. now there’s a date on it, so i just have to be ready.

added a draw unit today. you draw with your finger or mouse and the shape becomes control data for the sound — a curve you paint instead of a knob you turn. visual feedback while you draw, and it feeds the same deterministic engine, so the gesture becomes part of the seed’s world.

it’s the first part of the corruptor that isn’t a number or a toggle. everything else is precise, reproducible, cold. this one is a smudge you made with your hand. i like that tension — a fully deterministic machine with one input that’s just… whatever you scribbled.

spent the day making presets, which turned out to be the most fun part of the whole thing. every seed that made me stop and go “what the hell is that” got saved and named.

the list so far: GAS LEAK. MORGUE FREEZER. SIREN GRAVEYARD. WELDING ANGEL. PISTON CULT. DRILL SERMON. ELEVATOR TO HELL. IRON CHOIR. CONVEYOR HELL. STEEL LUNG. ACID FURNACE. STROBE CRUSHER. SPECTRAL CATHEDRAL. BLACK TIDE.

naming a sound is honestly half of making it. once “morgue freezer” has a name it becomes a real place you can go back to. and the preset itself is nothing — just a seed plus a few settings, tiny — but it reopens the exact same monster every single time.

turned a year older today. spent it in voronezh, nothing big going on. went to ddx in the morning, then sat at surf coffee by the lake for a while, and later had to catch a bus back to my town.

that was the whole day. no plans, no party, just a chill one. honestly that’s about what i wanted.

sound was solid, so i spent a day on how it FEELS. a tool you don’t want to touch just doesn’t get used, doesn’t matter how good it is under the hood.

put a real 3D logo in the corner — red extruded svg through three.js — and a barely-there dread mass floating behind everything, screen-blended, opacity like 0.14. you feel it more than you see it. added a reactive layer on top: a background spectrum field, a CRT oscilloscope, and every module got its own tiny dot-matrix display that shows what it’s doing to the sound. the display only moves when the module is on and something’s actually playing. dead when dead.

nobody warns you that the hard part of a sound tool isn’t the sound. it’s making the thing you HEAR be exactly the thing you EXPORT. spent a whole night just on that.

problem one: strudel’s reverb tail won’t die. calling hush() kills the notes but the reverb keeps ringing, and you can’t reach in and reset the convolver from the outside. so every time i switched a preset or rerolled, the old ghost kept bleeding into the new sound. fix — a hard switch that fully stops the audio graph before starting the next thing. in corrupted mode it’s perfect, the rendered buffer just gets cut instantly. in live mode the dry sound starts clean but strudel’s reverb still rings a little. that one i had to just accept — it’s the engine’s limit, not mine.

ok, the flagship. i call it the DC-77 DATA CORRUPTOR. the whole idea: a self-contained generator of collapsing sound, and everything — everything — is deterministic from one 8-digit hex seed. same seed = same sound, every time. type A7770077 and you get the exact same monster back, forever.

you pick a SOURCE: SHOT (a single hit), LOOP (a rhythm) or DRONE (a held tone). register, number of notes, and a ⌁ button that rerolls just the melody without touching the rhythm. then you run it through two racks.

been building a thing. browser-based industrial techno generators, all running on strudel (the live-coding engine, superdough under the hood). no daw, no plugins, no install — you open a page, hit play, it makes sound.

why. because i kept wanting weird raw material and didn’t want to open ableton every single time just to sketch a texture. so i put the sound sources in the browser instead. a kick synth with a pitch envelope and distortion, a rumble unit with four ways to grow the tail (reverb / decay / delay / sub), and a long atmosphere unit — chord drone with slow filter and phaser drift, the kind of thing that just breathes for two minutes and goes nowhere on purpose.

over the past few weeks i’ve been recording sets nonstop. put together mixes for MASS, Ego Death, and Monasterio — three very different vibes, three different approaches. feels good to have that kind of range on tape.

then on april 8th, wednesday, i went to an audition at Park Kultury — a local bar / club. played a 20-minute set. short format, no room to warm up, you either hit it or you don’t. the kind of thing where you walk in, plug in, and prove something in real time.

fifth session. hypnotic techno, percussion, and generative logic.

this is why you start with kicks and drums, not with synths. once the atmosphere is there, building the groove under it is genuinely harder. when the kick and percussion come first, everything else — synth patterns, note ranges, overall mood — starts to suggest itself. the tonality of a track often becomes obvious only after the drums are locked.

for percussion, two sources: ready loops and generated material. ready loops: pick something that gives the track its character — could be almost pure noise, could be textured metal. right away: low-cut EQ, ducking under the kick, clip or clipper if needed, level match by peak. in a good pack the samples are already treated, clipping is minimal. if you pull random material, run the clipper first before anything else.

fourth session. sample pack organization, rumbles, and a phase trap.

first thing: leveling the whole kick pack with a clipper. find the quietest usable kick, use the clipper to just catch the peak — barely touching the top. then match the rest to that ceiling. two to three decibels spread max. don’t cut too hard or you kill the transient character.

when saving samples: always crop the clip before dragging it out. serum adds a tiny silence at the start when playing samples internally — if you save without cropping, that silence ends up in the sample pack and causes misalignment later. crop, then save.

third session. deep into kicks and stereo.

spent a lot of time on utility — putting it at the end of every kick channel to control the stereo width. kick 3 was missing it, the other two had it. you hear the difference more in headphones than monitors. MSE explained why: headphones are better for sensing stereo, reverb, space. monitors are better for catching frequency balance — how the kick sits against the bass, how the low end actually behaves. you use both, but for different things.

second session. all about kicks.

MSE showed how to build one from scratch in serum 2 — sine wave, pitch drop via LFO in envelope mode, layered with a few more LFOs at different speeds to shape the attack. one for the body, one for the punch, one for the sharp click on top. also learned why phase matters: random phase off, phase at zero — so every hit starts the same and the speaker pushes forward from the first sample.

got into monasterio academy. first session with MSE — mostly intro, just talked.

he walked me through how the course works: 16 lessons, 2 months, 2 times a week. we’ll go through kicks, percussion, hats, scenes, live transitions, fx rack — step by step building an actual hybrid live set. that’s the final: a real performance.

he showed me his live in ableton. how scenes switch, how to keep levels from jumping when you move between them, how to think about frequencies not as mixing but as sound design from the start. a lot of things clicked that i was doing wrong without knowing.

built a unicode glitch text generator. paste it anywhere — telegram, notes, whatever you’re typing in.

ten corruption levels, different generation modes, optional geometry. pick chaos and watch it fall apart.

→ GLITCH ENGINE

recorded SHUM 004 today. had to redo it 4-5 times but it was a good training session overall — caught a lot of mistakes i didn’t know i was making. fixed them as i went. the final take came out solid.

→ SHUM 004

Party stil is prepping their live for Valentine’s Day at my house.

quite cool to see such things for the first time lol

https://www.instagram.com/party_stil?igsh=MTd1cWszemk4ODJ5dg==

my first ever recorded set. industrial techno freestyle, raw and experimental. recorded at home about a month ago, before my debut in Kazakhstan. the video is not synced with the audio — iphone had a setting messing with fps, making it dynamic and breaking the timing. just documenting the journey.

Metamorphosis is my debut industrial techno track, inspired by Kafka’s The Metamorphosis.

The intro and first section mirror the moment of transformation and alienation, while the ending closes the circle, following the structure of the book.

The track is built on pressure, repetition, and a slow shift of inner state, stripped of decoration.

Mastering and selected mix guidance by MSE (Main Sniffer Engineer).

u can text me on insta and i will send u the link if u wanna listen.

currently gonna send it to some labels.

How to stop being overstimulated: simple strategy
lately i am trying to be more present, otherwise i feel like i am not enjoying my life.

so vladilen minin is basically laying off most of result-school.

i work there btw.

fun times.

second week here was just me grinding at my friend’s place.

finally managed to develop the concept for an advanced learning platform that’s been sitting in my head for a while. feels good to get it out of the abstract phase and into something concrete.

sometimes you just need a change of environment to actually think.

finally a bar with good electronic music was opened in my town.

you can see a local dj lestay on the left prepping lol

спекулянт

finally i got a chance to get the fk out of my town to visit SPB again.

Will play with guys AAARONE and AVILKINADI

Some pics from my performance, testing if album and post pics synchronize lol

Final resting before releasing my unfinished website, it will never be finished (っ◕‿◕)っ

such a cool place, looks like an American high school or UC3M’s getafe campus in Madrid.

much better space to work or study, very thought provoking lol

ㄟ(￣▽￣ㄟ)

I am just testing my website functionality sitting in the mall called ХанШатыр in Astana.

I hope by the end of the trip i finally deploy this website

I AM TESTING LOCATION SCRIPT ON MY WEBSITE

FINALLY I HAVE CREATED FIRST VERSION ON MY WEBSITE.

SITTING IN TURKEY IN TITANITC DELUXE WHILE WRITING THINS