02:17 AM — The Terminal Is the Only Honest Place I Know
The timestamp in the corner of my monitor reads 02:17 AM.
The room is dark except for the cold blue glow of the screen, and the only sound is the low mechanical hum of my CPU fan spinning at a cadence that tells me something computationally expensive is happening. Which it is. Something is always computationally expensive at this hour.
I’m not watching anything. I’m not scrolling through anything. I am staring at a traceback error that has been defeating me for the last forty-five minutes, buried somewhere in the backward pass of a custom backpropagation engine that I wrote myself — from scratch, in pure Python, and I mean pure. No PyTorch. No TensorFlow. No automatic differentiation library of any kind. Just NumPy for the CPU operations, CuPy for the GPU, and whatever remains of my sanity at two in the morning.
The error is a shape mismatch between a (512, 256) weight gradient and a (256, 512) upstream delta. Which sounds simple. Which is not simple. In a hand-written backprop engine where you are manually tracking every tensor dimension through every layer of a network, a single transposed axis anywhere in the chain can silently corrupt your entire gradient flow — producing a loss curve that descends confidently in completely the wrong direction for twenty epochs before you realize what happened. I know this because it happened to me last week.
I have four browser tabs open. The NumPy broadcasting documentation. A 2019 paper on memory-efficient backpropagation from a research group at MIT. A GitHub issue thread where someone hit an almost identical CuPy kernel bug in 2022 and the resolution was buried in a comment chain on page three that nobody reads. And the fourth tab is a blank scratch document where I’ve been writing out the math by hand — the actual chain rule derivations, layer by layer — trying to locate the exact point where my tensor’s axes go wrong.
There is a glass of water on my desk that I have been meaning to drink for approximately two hours. It has long since achieved room temperature.
And then — there. I find it. A single axis inversion in my dense layer’s weight gradient calculation. A .T that was supposed to be there and wasn’t. One character. I add it, save the file, re-run the training script, and watch the loss start its descent in the way that makes you feel, briefly, like gravity is personally working in your favor.
The feeling when it works is not something I can describe to someone who hasn’t experienced it. It’s not excitement exactly. It’s more like the particular satisfaction of a system finally behaving according to the laws you wrote for it. I built the physics of this thing. I decided what the gradient would do. And now it’s doing it.
I’ve been calling this small project AINO. This is short for Aino is Neural Operation. The code is entirely public — it lives on GitHub and on PyPI, installable with a single command, every line of it open to inspection. No magic functions. No black boxes.
I close my laptop at 3:34 AM.
Exactly four hours and twenty-six minutes later, I am sitting in a wooden chair under flickering fluorescent lights, watching a teacher write a system of two linear equations on a whiteboard.
2x + 3y = 12
x − y = 1
“Find the value of x and y using the elimination method.”
I first solved this type of problem in 2022. I was in eighth grade. I was thirteen years old. I don’t say this to be cruel to anyone in that classroom. I say it because the specific cruelty of the contrast is worth naming clearly: the night before, I was computing partial derivatives through six layers of a neural network — chain rule applied recursively, tensor dimensions tracked manually, gradient flow debugged by hand. And now I am being asked, in all seriousness, to eliminate a variable from a system of two unknowns for the fourth year in a row. The same two variables. The same elimination method. The same format of the same problem I have solved dozens of times since the eighth grade.
The irony is not lost on me. Linear algebra is not my enemy — it is practically my bestfriend. I use matrix operations every single night. The difference between what I do at 2 AM and what is being taught on that whiteboard is not a difference in subject matter. It is a difference in depth — the difference between a paddling pool and the open ocean. SPLDV is the surface. What I actually need — eigendecomposition, gradient spaces, the linear algebraic foundations of how a neural network learns — lives somewhere the curriculum has never bothered to go.
I have to do this while operating on four and a half hours of sleep and the lingering intellectual residue of a problem that required everything my brain has.
I don’t fall asleep in that classroom because I’m lazy. I want to make that absolutely clear. The person who stayed awake until 3:30 AM debugging CuPy kernels is not a lazy person. I fall into a kind of cognitive flatline — because there is simply nothing in that room asking anything of me. No friction. No challenge. No moment where the material meets my current level and creates the productive resistance that real learning requires. My brain, which just spent hours working at the very edge of its capacity, arrives in a room where there is no edge. And so it idles. And I stare at the whiteboard with the specific thousand-yard expression of someone who is technically present but has achieved a state of almost meditative disconnection from the proceedings.
This is the daily reality. This is the gap I live in. And the gap is not a minor inconvenience — it is a kind of slow structural damage, compounding quietly, eroding the relationship between school and meaning, between institution and trust.
The Irony Baked Into the Bones of Vocational Education
Let me be precise about what I’m critiquing here, because I want to be fair and I want to be specific. I am not making a general argument against education. I am someone who cannot stop educating himself — this should be obvious by now. What I’m living inside is not education in the meaningful sense. It is the administrative ghost of education, still haunting the building long after the substance that originally justified it has moved on.
Vocational High School. The word “vocational” is doing significant work in that name. The implied contract when you enroll is clear: this is the practical track. This is where theory meets application. This is where you graduate with skills that the industry can actually use. That’s the promise that distinguishes SMK from SMA. That’s the value proposition.
Here is what a meaningful fraction of my actual school week looks like.
General subjects — mandatory national curriculum requirements that have no meaningful relationship to the IT vocational track I enrolled in — consume a significant portion of every week. And not just generically. Let me be specific, because the specificity is where the absurdity lives.
Indonesian Language, where we spend weeks analyzing hikayat — classical Malay folk tales written in a literary register that nobody has spoken in four centuries. The hikayat are not without beauty. As artifacts of a literary tradition, they are worth knowing about. But we are not studying them as cultural artifacts with context and depth. We are parsing sentence structures and answering comprehension questions about princes and magical kingdoms, assessed in the same format that I imagine has been used since my parents were in school. There is no discussion of how narrative structure works across forms, no connection to contemporary writing, no examination of why these stories mattered and what they displaced. Just the text. Just the questions. Just the waiting for the bell. Meanwhile, I have never once been taught how to stand in front of a room and articulate a technical idea. I have never been given a framework for how to write documentation that another engineer could actually use. Public speaking. Technical communication. The skills that determine whether your work gets adopted or ignored. These are the things the industry will judge me on the moment I walk through its door — and they have never appeared on a single syllabus.
Math, which at the SMK level cycles back, reliably, to SPLDV. I want to be very careful here, because I genuinely believe mathematical literacy is foundational and that linear algebra in particular is critical to everything I do. I derive gradients that live inside linear algebraic structures every single night. But SPLDV at the level it’s being taught — substitution method, elimination method, two variables, find x and y — is something I mastered in the eighth grade. The connection between that material and the actual mathematical depth I need — gradient descent, eigenvector decomposition, the calculus of attention mechanisms — is never made. The curriculum reaches just far enough to technically cover “linear equations” and stops before anything interesting happens. It is the mathematical equivalent of teaching someone the alphabet and calling them literate.
PPKN — Pendidikan Pancasila dan Kewarganegaraan, Civics — deserves its own paragraph, because it represents a particular category of educational dishonesty that I find genuinely difficult to sit through. The subject teaches us about national values, civic responsibility, the principles of democracy, the architecture of a just state. In theory, this is important. A citizenry that understands the philosophical foundations of governance and its own rights within that governance is a healthier citizenry. I believe this.
But there is a specific experience of sitting in a PPKN class and being asked to recite the ideal functions of state institutions that you know, from the news you read every day, are not functioning the way the textbook describes. There is a specific cognitive dissonance of memorizing the principles of transparent and accountable governance for an exam while simultaneously knowing that the gap between those principles and the practiced reality is enormous, well-documented, and largely not discussed anywhere near the classroom. PPKN, as I have experienced it, does not teach you to think critically about how governance actually works and where it fails and why. It teaches you the official story. It teaches you to perform civic understanding. It teaches you, in the most charitable interpretation, to pretend to be a healthy citizen in a country that is still working through very serious structural illness. And the pretending, repeated often enough, starts to feel like the point.
And so on. And so on. Week after week, subject after subject, the pattern holds: material calibrated for a student who is not me, delivered in a format designed for compliance rather than comprehension, assessed in ways that measure retention rather than understanding.
The result: I sit there. I endure. I clock out mentally. I count the minutes.
And this is where the invisible damage accumulates. It’s not just boredom — boredom is manageable. It’s that spending extended periods in an environment that asks nothing of you starts to erode your belief that school is a serious place. When an institution consistently signals that it doesn’t know who you are, you begin to disengage from it not as individual teachers or individual classes, but as a premise. The institution stops feeling like a real environment — it starts feeling like an obligation to metabolize before you can get back to the real work.
Most of my teachers are genuinely trying. I want to say that clearly. They are working within a system that gives them very little room to deviate from the assigned syllabus. The problem is not the people. The problem is the structure, the incentives, and the inertia.
And the inertia is staggering when you actually look at it from the inside.
Consider what the technology landscape has done in the time I’ve been in school. We have gone from BERT as the dominant NLP paradigm to ChatGPT 5.3’s emergent capabilities to the Transformer architecture eating every other domain to LLMs becoming commodity infrastructure to multimodal models to reasoning models that use extended chain-of-thought to solve problems that the generation before them couldn’t touch. This is not incremental change. This is the field repeatedly turning over its own foundational assumptions and rebuilding from a new starting point. Every practitioner working in AI has had to develop not just new skills, but new mental models.
None of this has penetrated the curriculum. The curriculum is still describing the world of 2015, with minor cosmetic updates. In a field where a decade is geological time — where the tools I’m using today didn’t exist five years ago — this is not a minor update problem. This is a foundational philosophical failure: a curriculum that was supposed to prepare students for an industry cannot prepare students for an industry it has stopped tracking.
I don’t hate this school. What I feel is more complicated than hate. I feel the specific frustration of someone who can see clearly what this place could be, and what it actually is, and the distance between those two things — and who has to show up in the actual version every single morning.
The Double Burden Nobody Talks About
There is one more thing about the SMK experience that I have been trying to find the right words for, and I think I finally have them.
SMK students are expected to absorb the full weight of the general education curriculum — the same hikayat, the same SPLDV, the same civic performances as SMA students — and simultaneously master the vocational competencies that are theoretically the entire reason the SMK track exists. We are expected to be fluent in IT fundamentals, complete industry-aligned practical projects, hold our own against SMA graduates in standardized assessments of general knowledge, and somehow also do all of this while preparing for a job market that will judge us by different and harsher criteria than our SMA peers.
In other words: we are expected to be as academically complete as SMA students while also being as technically prepared as industry trainees. Both. Simultaneously. With the same hours in the day.
And when we inevitably struggle to do both perfectly — when the cognitive bandwidth required to memorize hikayat vocabulary for tomorrow’s exam eats into the hours I need to actually understand gradient computation — the implicit diagnosis is that we are less serious, less capable, less academically rigorous than SMA students. The failure is attributed to us. Not to the structure that created the impossible double bind in the first place.
This framing infuriates me in a very specific way.
The students who choose SMK over SMA are not, as a category, students who couldn’t hack it in the academic track. Many of us chose SMK precisely because we are serious — serious enough about a particular domain to want to spend our formative years going deep into it rather than wide across everything. That is not a failure of ambition. That is a clarity of ambition that the system then systematically punishes by loading us with everything we were theoretically choosing not to prioritize when we picked the vocational path.
I chose IT because I want to build things. I want to understand how systems work at the level where you can construct them yourself. I want to write something that good for people, not fill in worksheets. The choice to be here was an act of intentionality — and the system’s response to that intentionality is to spend a significant fraction of my time on material that has nothing to do with why I made the choice.
There is a version of this that I can accept philosophically. A broad general education has value. Exposure to language and mathematics and civic reasoning makes better engineers, not just better test-takers. I genuinely believe this. But there is a difference between a broad general education delivered with depth and integration — one that connects the foundations of mathematics to the calculus of optimization, that uses Indonesian literary tradition to teach structural thinking about narrative and argument, that uses civics to develop critical thinking about institutions — and a broad general education delivered as a compliance exercise. What we have is the latter. The form without the substance. The checkboxes without the education.
We are expected to be perfect on the SMA terms. We are then evaluated by industry on the SMK terms. And we are given full preparation for neither.
That is the trap. And I think it needs to be named clearly, because the students who struggle inside it often interpret their own difficulty as personal failure rather than as a rational response to an irrational structure.
Why University Is Not My Path Right Now — And Why I’ve Thought About This Carefully
Now we need to talk about the degree question, because this is where people will push back hardest. “Just get the credential. Then do whatever you want.” I have heard this argument many times. I take it seriously enough to actually engage with it rather than dismissing it, because there are real cases where it is the right advice.
For me, right now, it is not the right advice. Here is why.
The time cost doesn’t add up the way people assume it does. An S1 degree in Indonesian takes a minimum of four years. Four years in a field where the foundational tools and paradigms turn over every twelve to eighteen months. The AI skills that would be central to a 2026 CS curriculum — attention mechanisms, transformer architectures, fine-tuning foundation models, building on top of the LLM stack — were either nascent or nonexistent when the professors who would teach that curriculum were being trained. The gap between academic CS curriculum and industry state-of-the-art in ML/AI is a known, documented, publicly acknowledged problem that practitioners at world-class institutions openly complain about. If the curriculum can’t keep pace at Stanford, I have limited confidence in how the material will age at the average Indonesian university. By the time I graduated in 2030, I would be entering the field with skills that were state-of-the-art in 2026 — which, in this domain, is a significant disadvantage.
The financial logic doesn’t hold for everyone. Tuition, living expenses, textbooks, and four years of opportunity cost add up to a number that, for many Indonesian students from non-wealthy backgrounds, involves real family sacrifice. The implicit bet embedded in that sacrifice is that the credential will pay off in access and compensation. That bet is increasingly uncertain as the AI job market — globally, and slowly in Indonesia as well — shifts toward skills-based evaluation at the companies doing the most interesting work. The same resources invested in compute credits, quality online courses from practitioners who are currently in the industry, and hardware for real projects would compound differently. Not universally better — I want to be honest about that. But for my specific learning style and my specific situation, I genuinely believe they compound better.
The self-directed learning loop is not a compromise. It is a feature. Here is something about university that doesn’t get said enough: it outsources your intellectual accountability. Someone else determines what you learn, in what sequence, evaluated by what method, on what timeline. For some people — many people — this structure is exactly what they need to make forward progress. I understand and respect that. But I know with a fairly high degree of self-awareness that I am not that person. I learned backpropagation not because it was assigned, but because I needed it to fix a bug in my own code. I learned CUDA memory management not from a textbook, but from watching my own training loop choke on GPU transfers and refusing to accept not understanding why. The problem-first, curiosity-driven learning loop I’ve developed over years of working this way is, I believe, a more efficient learning engine for me than the assignment-first, deadline-driven loop that university provides. AINO is proof of concept for this belief. Eighty-eight percent accuracy on an NLP task, achieved by a student with no formal higher education in machine learning, using a code he built himself from mathematical first principles — that is a data point.
To be completely transparent: I am not closing the door on formal education forever. If the right program, with the right community and resources and intellectual culture, appeared and made sense for where I am — I would consider it genuinely. Formal education is not the enemy. Credential gatekeeping is the enemy. Curriculum stagnation is the enemy. The unexamined assumption that the only legitimate path to expertise is institutional is the enemy. Those are separable from the idea of education itself, and it matters to keep them separate.
The Wall — And the Specific Geometry of Its Cruelty
And now we arrive at the part I don’t like to talk about, because talking about it requires admitting that all the technical achievement in the world runs directly into a structure I didn’t build and cannot dismantle on my own schedule.
I want to take you inside the specific experience of opening LinkedIn as an SMK student who builds deep learning frameworks from scratch. It is an experience with a very particular emotional texture that I’ve learned to recognize.
It usually happens in the evening. The productive energy of a coding session has faded. The cursor is blinking in an empty terminal and I’m not sure what to work on next. I open LinkedIn — partly out of genuine curiosity about what the industry looks like right now, partly out of a masochistic impulse I haven’t fully outgrown to test reality against my ambitions. I start scrolling through AI/ML job postings. And for a few seconds, reading the job titles — Machine Learning Engineer, AI Research Scientist, Deep Learning Developer, Applied AI Engineer — there is a pulse of recognition. These are my people. These are the problems I think about. These are the roles that, in some technical sense, I am already doing a version of every single night.
Then I read the requirements.
“Bachelor’s degree in Computer Science, Mathematics, Statistics, or a related engineering discipline required. Master’s or PhD strongly preferred. Minimum 3 years of industry experience.”
And I feel something I can only describe as a specific kind of deflation. Not devastation — I’ve metabolized this enough times that it doesn’t floor me anymore — but a quiet, heavy settling. Like a door closing in slow motion. Like a room you’ve been working toward turns out to have a lock you don’t have the key for, and the lock doesn’t care how hard you can pick.
Let me describe the mechanism that makes this painful in a way that ordinary rejection wouldn’t be.
It’s not that I think I deserve every job I see posted. I’m seventeen. I know there are engineers with years of industry experience who are better at this work than I am in ways that experience genuinely provides — and I have respect for that gap. What stings is the specific geometry of the barrier. It is not a skills barrier. It is a credential barrier. The question most of these postings are actually asking is not “Can you do this work?” — it’s “Do you have the paperwork that suggests you might be able to do this work?” Those are radically different questions with radically different implications for who gets access.
Here is what exists on the other side of my GitHub link: a deep learning smoll project written in pure Python, with no PyTorch or TensorFlow dependency, where every gradient is computed manually. A training pipeline that I optimized from 32 minutes down to 19 seconds by rethinking GPU memory transfer patterns at the kernel level. An NLP classification model achieving 88% accuracy, trained on a framework I derived from mathematical first principles and published to PyPI under a name that means something to me. This is a real artifact of real technical work. Anyone in the world can examine the source code and form an independent opinion about the quality of the thinking behind it.
That artifact is invisible to an Applicant Tracking System.
An ATS is a blunt instrument. It is designed to reduce a large pool of applications to a manageable shortlist by filtering for specific checkbox fields: Degree type. Recognized institution. Years of experience. What it cannot do — what it is architecturally incapable of doing — is follow a GitHub link and understand what it finds there. It cannot read a benchmark table and understand the engineering decision-making behind a 100x performance improvement. It cannot evaluate the quality of a mathematical derivation in a code comment. It looks for the checkbox. I don’t have the checkbox.
So here is the probable reality of my job search as it stands today: if I submitted applications to fifty AI/ML roles in the Indonesian market or internationally, a realistic estimate is that fewer than five would survive automated screening. Not because I lack the skills the roles require. Because I lack the checkbox.
The signal and the thing it’s supposed to signal have been decoupled, and the system has not noticed.
This creates an identity crisis that I think about more than I’d like to admit.
I am too advanced for my environment. At school, I am occupying a classroom that cannot meet me where I am technically. The curriculum is not designed for people who spend their nights working at the edge of what they understand. I say this without arrogance — it is a straightforward description of a mismatch.
At the same time, I am invisible to the industry. The companies doing the most interesting AI work operate hiring pipelines that filter me out before a human sees my work. I don’t have the credential, so I don’t get the interview, so I can’t demonstrate the capability that the credential is supposed to signal.
Trapped above the floor and below the ceiling. Too far ahead for here, not yet legible to there.
This is the peculiar loneliness of the self-taught non-traditional learner who is genuinely serious. The narrative we usually receive is one of two extremes: the triumphant “dropped out and built a unicorn” story that requires a specific combination of luck, timing, and network that most people don’t have access to — or the cautionary “you need the degree” story that flattens every individual path into a single recommended route. The lived reality for most of us is messier and more honest than either. It’s characterized by long stretches of building things in the dark, betting on a future where the work will eventually become visible to someone who can do something about it.
Treating Anxiety Like a Debugging Problem
Here is something I’ve been doing with the anxiety, because it turns out that anxiety is actually somewhat tractable if you approach it like a debugging problem rather than a weather system you have to wait out.
When I examine the source of my industry anxiety carefully — really trace it back to its root cause — it’s not fundamentally about the credential itself. The credential is a proxy. What the anxiety is actually about is legibility: the question of whether I will ever be visible to the people who could recognize the value of what I’m building. The ATS problem is a legibility problem. The degree requirement is a legibility problem. The challenge, framed properly, becomes: how do you make yourself legible to a system that wasn’t designed to read you?
One answer, which I’ve considered and rejected, is to become legible on the system’s own terms — to pursue the degree specifically as a signaling mechanism, not because of the learning it would provide, but because of what the checkbox tells the algorithm. This is a coherent strategy. I understand why people choose it. But it has a cost: four years and a significant financial commitment, invested primarily in acquiring a signal rather than a capability, in a field where capability is what you’re evaluated on once you get through the door. If the entire game were about getting through the door, that might make sense. But the game doesn’t end at the door. And spending four years on a signal rather than a capability means arriving at the door four years behind.
The answer I’m actually pursuing is to build alternative legibility. To create artifacts of work that are so clear in their quality and specificity that they function as credentials within the communities and pipelines where humans — not algorithms — are doing the evaluation.
This is harder. It is slower. It requires betting that I can become visible enough, to the right people, in enough time, that the absence of a formal credential doesn’t permanently close the doors I need open. It requires holding a nuanced view of the tech industry — acknowledging both the real meritocratic elements and the very real ways that meritocracy is unevenly distributed, gamed, and mythologized.
But it is the path I have. And I am going to run it as hard as I can.
The Resolution — What Defiance Actually Looks Like at 3 AM
Let me tell you what I’m building. Not as a wish list. As a plan with a logic behind it.
**AINO. ** This is the anchor of everything else. The framework exists, it is installable, it is documented, its benchmarks are reproducible. If you are a hiring manager or a fellow engineer who actually cares about what someone can do — not what paper they hold — you can read the source code and form an independent opinion. That is the point. I cannot control whether an ATS reads my GitHub link. I can control whether what’s on the other end of that link is worth reading.
The technical writing is continuous with the technical work. This essay is not a separate activity from my engineering. It is the same cognitive process expressed in a different medium. Engineers who can think clearly about complex systems and communicate that thinking to other people are — not universally, but generally — engineers who understand what they’re doing more deeply than engineers who cannot. Writing forces precision. Precision requires understanding. I am investing in both the depth and the articulation, because both compound, and both are visible in different contexts.
Open source is one of the genuinely meritocratic spaces. Beyond AINO itself, I am mapping the ML open source ecosystem for projects where a contribution from a non-credentialed self-taught developer from Indonesia is evaluated purely on its technical merit — because that is how open source works at its best. A good pull request is a good pull request. The commit author’s degree field is not visible in the diff. There are spaces in this industry where what you did is legible independent of where you studied, and I intend to become a known entity in those spaces.
The community is infrastructure. The ML communities on Twitter/X, Discord, Hugging Face, GitHub — these are full of working engineers, researchers, and practitioners who make hiring decisions, recommend candidates, and build products. Becoming a recognized presence in those communities — someone whose work people know and whose thinking people engage with — is a form of career building that has no credential prerequisite. It requires showing up consistently with good work and genuine intellectual curiosity. Both of those I have in abundance.
I want to be honest about what I don’t know. I don’t know if this path will get me where I want to be on the timeline I’m hoping for. I don’t know whether the credential barrier in the Indonesian market will relax fast enough to matter for me during the years I’m trying to break in. I don’t know if AINO is good enough yet — I think it demonstrates real capability, but “good enough” in a competitive market is a function of what you’re being compared to, and I don’t have full visibility into that comparison.
What I know is this: the alternative path — waiting, complying, deferring to a system that doesn’t see me, hoping that patience will be rewarded by an institution that hasn’t updated its curriculum in a decade — has no feedback loop. The path I’m on has a feedback loop. I build something, I publish it, I get responses, I iterate. My output affects my trajectory. That feedback loop is worth more to me than the certainty of waiting.
A Letter to the Gap
If you are somewhere in this gap — and I know you’re out there, because this gap is not unique to Indonesia, or to SMK, or to machine learning — I want to say something directly to you.
The gap is real. Do not let anyone flatten it into impatience or arrogance. The people who tell you to just be patient, that the system is basically fair and will recognize you eventually, are often people who have already made it through the system and need to believe that it is basically good. Some of them mean well. The system is still structurally resistant to people who don’t fit its pattern, and that resistance is real and you will feel it.
At the same time: the gap is not a permanent condition. It is a current state. Systems change — slowly, imperfectly, in the direction of the evidence that accumulates whether the system acknowledges it or not. The history of this industry is full of people who built things while they were invisible to the credential apparatus of their moment — who kept going because the feedback loop of the work itself was enough — and who eventually became impossible for the system to ignore.
I don’t know yet if I’ll be one of those people. I am betting my time and sleep on it. But I’m not betting on arrogance. I’m betting on evidence. The loss function is actually descending. The gradients are correctly computed. The training time is actually 19 seconds. The accuracy is actually 88. These are not aspirations. They are measurements. And measurements, unlike credentials, cannot be revoked.
Keep building. The proof of work is the only currency that never inflates.
Back to the Wooden Chair
Tomorrow morning the alarm will go off at six forty-five. I will have slept approximately five hours. I will drink cold coffee standing over the kitchen sink. I will go to school. I will sit in the wooden chair. The whiteboard will fill with things I have known for years, and I will practice the art of being physically present while my mind is somewhere more useful — maybe thinking about the attention mechanism I want to implement next, maybe planning the documentation for AINO’s next release, maybe just quietly recovering from the night before.
And when school ends, I will come home. The laptop will open. The terminal will come back to life.
The loss will descend.
The gradient knows where to go.
This is what I am. This is what I’m building. The SPLDV equations are still written on the whiteboard somewhere in that school. The framework is still on PyPI, with its benchmarks and its source code and its reproducible results, waiting to be found by anyone who is looking.
Both of these things are true. Only one of them is building toward anything.
I know which one I’m living for.
If this resonated with you — whether you’re an SMK student, a self-taught engineer, someone navigating a non-traditional path in any country or field — I want to hear from you. Not because I have the answers, but because I think we need to build the map of this territory together, in public, where it can actually be useful to the next person trying to navigate it.
Find AINO on PyPI. Find me on GitHub. If you see something broken in the code, open an issue. If you see something worth building on, fork it and make it better. That’s the whole game — and anyone can play it.
This post is strictly a personal opinion and rant from the perspective of a vocational high school student named Alpha regarding the gap between the formal education curriculum and the reality of the tech industry, serving not as an attack on any school or institution, but rather as a genuine documentation of a self-taught learning journey.