Nearly Three Million Streams, One Ban, and No Spin: The Whole Story of Daryl Belhida, the Builder at NuDoors

Here is the strangest line item you will find in any builder's portfolio this year. A developer in the Philippines released music through a distributor called Amuse, and the music did what almost no independent music ever does: it found an audience. Nearly 3 million streams, spread across Spotify, Apple Music, YouTube Music, Deezer, and YouTube. If the story stopped there, it would be a victory lap, the kind of number a person puts at the top of a profile in bold type.

The story does not stop there, and the reason this post exists is that the man himself refuses to let it stop there. The songs were AI covers. Later, the covers were removed, and his Amuse account was banned. That is the ending, and Daryl Belhida tells it himself, in his own profile, in the same breath as the triumph. No footnote, no quiet omission, no hoping you will not ask. Nearly 3 million streams, one ban, both numbers volunteered by the same person.

Sit with that for a second, because it is rarer than it looks. The internet is currently flooded with AI success stories, and almost every one of them has been edited. The revenue screenshot is real but cropped. The user count is real but counts everyone who ever clicked. The demo works, as long as you do not ask what happened when a real customer touched it. We have all learned to read these stories the way you read a fisherman's account of the one that got away: politely, and with a private discount applied to every number.

Daryl's profile applies the discount for you, in advance, at cost to himself. And here is the thing: the ban is not even the main part of his work. By day he is a full-time full-stack developer at a company called NuDoors, where he built an AI tool that designs your new kitchen from a photograph. On his own time he built a platform that turns a family photo into a printed coloring book and mails it to your door. His portfolio runs twelve projects deep, across web platforms, mobile apps, and connected hardware. The music was a side quest. But the side quest is where you can see the man's operating system most clearly, because the side quest is where he had something to hide, and he declined to hide it.

This is the whole story, told the way he tells it: the streams and the ban, the kitchens and the coloring books, the discipline underneath all of it. By the end, we will get to the question his profile quietly asks every builder and every person who hires one: in a market drowning in spun numbers, what is an unspun story worth?

A developer in Talibon, building for a cabinet company in America

Start with the geography, because the geography is the first surprise. Daryl Belhida lives in Talibon, Bohol, in the Philippines. He is a BSIT graduate, and in the E2 Discord, where his verified builder profile lives, he goes by Dyl. His day job is at NuDoors, an American cabinet refacing company. If you are picturing a freelancer picking up gigs between other gigs, correct that picture: he works there full-time, as a developer on the platform that sells and runs the business.

His stack is the modern full-stack standard, and he covers all of it. React and Next.js with TypeScript for the web. React Native when the product needs to live on a phone. Firebase and Node.js underneath. On top of that sits nearly four years of using AI for development and creative work, which, if you count backward, means he started building with these tools before most of the industry had finished arguing about whether they were a fad.

Inside the AI Pro applicant pool, the community of builders his profile belongs to, he holds two small distinctions that tell you a lot. He is one of only two people in the entire pool who has actually used Muse by Meta. And he was the first person in the pool to run a recurring task inside it: a cron job, a small scheduled program, that goes out and collects AI news, tools, and examples for website and course projects, then summarizes them, every Monday, whether he remembers to ask or not.

Hold onto that Monday cron job. It is a tiny detail, easy to skim past, and it is one of the most revealing things in the whole profile. Most people use AI the way they use a search engine: they show up, they ask a question, they leave. Daryl set his tools to work on a schedule, in the background, compounding quietly while he did other things. That instinct, automate the recurring thing and spend your attention on the judgment calls, turns out to be the pattern under everything else he builds.

He can start immediately, he is flexible on hours, he is comfortable working US hours, and he is happy to do a paid trial. Those are the logistics, and his AI Pro profile lists them plainly. But logistics are not why you would hire him, or why we are writing about him. The reason is a habit of mind that shows up in his code first, and in his storytelling later. It has four steps.

The discipline: investigate, test, implement, verify

Daryl's main coding agent is OpenAI Codex. Plenty of developers use a coding agent now; the interesting question is never which one, it is how. The sloppy way to use an agent is to describe what you want, accept whatever comes back, and ship it if the screen looks right. The agent becomes a very fast way to produce code nobody understands, including the person who asked for it.

His way, as his profile describes it, runs in a strict order: investigate an issue, run automated tests, implement the change, verify the result. Read that list slowly, because the order is the whole philosophy. Investigation comes before implementation, which means the agent's first job is to understand the problem, not to start typing. Tests come before the change is trusted. And verification comes last, as a separate step with its own name, which means "it compiled" and "it looks fine" do not count as done. Done means checked.

This is the unglamorous truth about working with AI agents that the highlight reels leave out: the agent is not the skill. The loop around the agent is the skill. Anyone can generate code now. The scarce abilities are knowing what to ask for, knowing how to tell whether you got it, and refusing to call something finished until it has survived an honest test. Daryl's loop is built out of exactly those abilities, in exactly that order.

He also builds applications that integrate OpenAI and Google Gemini directly, so the AI is not only in his toolbox, it is inside his products, facing customers. That double fluency matters more than it might seem. A developer who uses agents to build, and also ships AI features to real users, learns the failure modes from both sides: how the tools behave when you are the operator, and how they behave when your customer's evening depends on them.

There is a reason the verify step is the one everybody else skips, and it is worth naming. Generation has become nearly free, and generation feels like progress: screens fill with code, and the day looks productive. Verification feels like friction. It produces no new material, it takes time the schedule wants for building, and worst of all, it can tell you the work is wrong, which is information nobody enjoys receiving at 4 in the afternoon. So the industry default is quietly becoming: generate, glance, ship, and let the users do the verifying. You can see the results in every software category you touch. The developers who stand out in this environment are not the ones who generate the most. They are the ones whose loops contain an honest checkpoint that the work has to pass before it is allowed to call itself done. That checkpoint is a personality trait before it is a technique, and it is the same trait that later makes a man publish his own ban. A person who will not let an agent's work slide will not let his own story slide either.

Now notice something. When his profile sums up his character at the end, it describes his approach in almost the same four beats: investigate, implement, verify, report. The reporting step is the one most engineers quietly drop. Reporting means the result gets told straight, to whoever needs to hear it, including the parts that did not work. Keep that word in your pocket. We are going to need it when we get to the music, because the music story is where Daryl applied the reporting step to himself, in public, where it cost him something.

First, though, the day job, and the photograph that replaced a salesman.

nuVision and the death of the home-visit estimate

NuDoors sells cabinet refacing, which is one of those sensible, unglamorous ideas that turns out to be a small revolution for a household budget. When a kitchen is refaced, the customer keeps their existing cabinet boxes, the sturdy frames already hanging on the wall, and replaces what the eye actually sees: new custom doors, new drawer fronts, new hardware. Most kitchens are done in about three days, instead of the weeks a full replacement takes, with the dust, the dumpster, and the eating takeout in the living room that a full replacement involves.

Now think about how a company like that traditionally sells. The old way starts with a home visit. A salesperson drives out, walks into your kitchen, measures your cabinets, talks through styles, goes away, does arithmetic, and eventually produces an estimate. Every step of that chain costs the company money and costs you time and a certain amount of social pressure, because it is much harder to say no to a person standing in your kitchen than to a number on a screen. The estimate is the bottleneck of the whole business: slow, expensive, and wrapped in obligation.

Daryl built the thing that removes the bottleneck. It is called nuVision, and it is the AI piece of the NuDoors platform. You upload a photo of your kitchen. You explore cabinet styles and colors. The AI renders your new kitchen for you, and it estimates your cabinet sizes and your project pricing, without measuring a thing. The site sums up the pitch in six words: no appointment, no measuring, no pressure. The estimate a salesperson used to produce on a home visit, generated from a photo instead.

Here is the deeper pattern, and it is worth slowing down for, because it applies to far more than kitchens. A demo of nuVision would be the render: look, the AI redrew your kitchen with different doors. That is impressive for about nine seconds. A product is what happens around the render. Photo in. Styles explored. Sizes estimated. Price out. And then, the step that makes it a business instead of a toy: the lead is captured for the company behind it. Daryl did not build a picture generator. He built a complete loop that starts with a stranger's curiosity and ends with a priced project and a customer the business can follow up with. His profile calls it exactly that: a complete product loop, not a demo.

And nuVision does not float alone. At NuDoors he works on the platform that sells and runs the business: booking, lead capture, and admin reporting alongside the customer-facing tools, plus contractor tools in the mix. That surrounding machinery is the part of software nobody posts screenshots of, and it is the part that determines whether a company can actually operate. The AI visualizer is the door. The platform is the house. Daryl builds both, which is precisely the combination that the current moment rewards: the flashy AI surface, and the boring, load-bearing systems underneath it.

Step back and notice what the estimate really is in a business like this, because it generalizes to almost every service business on earth. The estimate is not paperwork. The estimate is the moment where curiosity becomes commitment, where a person daydreaming about a new kitchen becomes a customer with a price, a plan, and a reason to call back. Whoever controls that moment controls the sale, and the traditional process put it behind a scheduled visit, a measuring tape, and a stranger's afternoon. Moving it into a photograph does not just save time. It changes who the tool belongs to. The customer runs it at midnight, in pajamas, with zero social pressure, and the company still gets the lead. That is what an AI feature looks like when it is designed by someone thinking about the whole business instead of the demo: the technology disappears into a better arrangement of the transaction itself, and everybody in the transaction can feel the improvement.

LumiLabs: a coloring book that is secretly a commerce system

If nuVision is his day-job proof, LumiLabs is the independent one, and it might be the most instructive product in the whole portfolio. Luminant Labs, which presents itself as LumiLabs, sells personalized coloring books. You upload one photo of your family, your friends, or your pets. The AI draws a whole book of adventures from it. You preview every page before you pay. Then a printed book, a physical object, is mailed to your door.

The details are all there in the open, the way details tend to be with this particular builder. The first book is free to make. A book runs 24 pages with a cover. There are four art styles and hundreds of adventure scenarios to draw from. Printed books cost $19.99, with bulk pricing for events. Nothing about this list is accidental, and if you read it the way a builder reads it, you can watch a dozen small business decisions hiding inside what looks like a charming gift product.

Start with the free first book. The expensive, uncertain part of a personalized product is the leap of faith: will the AI actually turn my photo into something my kid will love? LumiLabs removes the leap. You make the book first, you look at every page, and only then does money enter the conversation. The preview is not a feature; it is the entire trust strategy, and it rhymes with everything else in this profile. Verify before you commit. Show the result, then ask.

Then look under the charm, because underneath it sits a real commerce system, and this is where most AI side projects quietly die. There are image-generation workflows turning one photo into a coherent book of adventures. There are payments. There are subscriptions. And there is print-on-demand fulfillment, which means physical books get produced and shipped without Daryl warehousing a single box. Each of those is a system that has to work while he is asleep, in the right order, with real money attached. A coloring book sounds like a weekend project. A coloring book with payments, subscriptions, and a printer mailing your grandmother's gift on time is a business.

This is the same pattern as nuVision wearing different clothes. In both cases, the AI moment, the render of the kitchen, the drawing of the book, is the small visible tip. The product is the loop around it: input, exploration, verification, price, capture, fulfillment. People who follow AI casually tend to ask, what can the model do? People like Daryl ask a colder and more useful question: what has to be true, end to end, before a stranger will pay for what the model does? His two flagship products are what that question looks like when somebody actually answers it.

It is also, his profile notes, the work he points to first when someone asks what he can do. Given that the portfolio contains twelve projects, the fact that the coloring books and the kitchen visualizer lead the list tells you what he values: not the cleverest thing he has built, but the most complete.

One more angle is worth appreciating, because it is easy to miss: the end of the LumiLabs loop is physical. A kitchen render lives on a screen and can be regenerated if it disappoints. A printed coloring book has to exist. Paper, printing, binding, postage, an address, a door it arrives at. The digital half of the business can be iterated at the speed of thought, but the physical half answers to the old world's rules, where a mistake is not a bug report, it is a ruined birthday gift. Builders who ship physical fulfillment learn a humility that pure software rarely teaches: the loop is not closed when your code says so. It is closed when the box lands. A platform that reliably lands boxes, built by one developer, on top of an AI generation pipeline, is a genuinely hard thing masquerading as a cute one, and the cuteness is doing commercial work too. Nobody brags to their friends about a well-architected subscription system. They absolutely show people the coloring book with their dog in it.

Twelve projects deep

The rest of the portfolio is worth a walk, because the spread itself is the story. Twelve projects, and they do not huddle in one safe category. There are web platforms and visualization tools built in React and Next.js with TypeScript, Firebase, and APIs. There are mobile applications in React Native, including Voizz, a communication app, and TimeCub, a parental control app, two products aimed at completely different humans with completely different worries. And then the portfolio leaves the browser entirely: connected systems, IoT and embedded work, data dashboards, hardware integration.

A generation ago, that range would have described a small agency. Today it describes one developer with a disciplined agent loop and nearly four years of practice at driving it. This is one of the quiet, enormous changes the tools have made: the cost of starting a project in an unfamiliar category has collapsed, so the bottleneck moves from "can I build this at all" to "can I understand the person it is for." Daryl's own summary of his background lands exactly there. Kitchen visualization, contractor tools, personalized coloring books, mobile apps, and connected hardware, and then his stated principle, which his profile quotes from him directly: whether writing an app or troubleshooting a system, he cares about the person on the other side of the screen.

Look at the two named mobile apps for a moment, because they share a hidden property. Voizz is a communication app, a product whose entire value is people trusting it with their conversations. TimeCub is a parental control app, a product whose entire value is parents trusting it around their children. Neither category forgives sloppiness, and neither can be faked with a good landing page: trust products either behave, every day, or they are deleted. A builder whose portfolio gravitates toward products people must rely on, kitchen estimates a company sells from, books a family pays for in advance, apps parents install on a child's phone, is telling you something about his center of gravity. He builds things that have to work when he is not in the room. That is a harder standard than things that merely have to demo well, and it is the same standard, once again, as a story that still works after you include the ending.

His creative work runs through the same machinery. He creates and edits videos through code, using ElevenLabs MCP for voiceovers and Veo for video generation, which means the video pipeline is itself a program he wrote, not a timeline he drags clips around on. The software builder and the media maker are the same person using the same method: set up a system, let it do the repetitive work, spend your own attention on judgment.

Which brings us, at last, to the music. Because the music was made with the same tools and the same energy, and it reached more people than everything else in this post combined. And then it ended the way it ended, and he told everybody.

The music, told whole

Here is the music story with nothing removed, in the order it happened, using only what Daryl himself reports.

It starts with a win. He made a jingle with Suno, the AI music tool, and the jingle won a college competition. Pause on that detail, because it frames everything after: this did not begin as a scheme. It began as a student making something with a new instrument and being told, by judges, that the something was good.

Then came the releases. He put music out through Amuse, a distributor, the kind of service that takes a recording and places it across the streaming world. And the music traveled. Nearly 3 million streams in total, across Spotify, Apple Music, YouTube Music, Deezer, and YouTube. To calibrate that number: the overwhelming majority of music released in any year, human or otherwise, is heard by almost no one. Nearly 3 million streams means nearly 3 million moments where a real person, somewhere, pressed play and stayed. Whatever else is true of this story, the audience was not imaginary.

The releases included AI covers. Notice the precision of that word in his telling, covers. He does not say "my music was caught up in a policy change" or "some tracks had issues," the kind of fog people reach for when a story turns. He names the category exactly: they were AI covers, recordings of existing songs, made with AI. That precision is the whole story in miniature. At every point where a vaguer phrase would have protected him, he chose the exact one.

And here is the ending, in his own telling: the AI covers were later removed, and the Amuse account was banned. Not one song quietly delisted while the rest of the catalog kept earning. The covers came down, and the account itself, the door through which all of it had been released, was closed.

His profile gives no further reason, and this post is not going to invent one. That restraint is deliberate, and it is exactly the point. Guessing at the distributor's reasoning, dressing the gap up as analysis, softening the ending with context nobody provided: that would be spin, and spin is the one thing this story is being used to argue against. What we know is what he reported. The covers were removed. The account was banned. He included both facts in the document whose entire purpose is to persuade people to trust him with their projects.

Consider what it would have cost him to leave them out. Nothing, visibly. The streams number alone would have gleamed on the profile, and essentially no reader would ever have gone looking for the ending. Consider, then, what including them costs: every skeptical reader now knows the worst fact in the file, at the exact moment they are deciding whether to believe the rest of it. He paid that cost voluntarily. Why would a person do that?

What the bad half of a story is worth

The answer is that the bad half is the receipt for the good half.

Think about how belief actually works when you read a stranger's claims. A number with no visible downside attached is just an assertion; you can admire it, but you cannot weigh it, because you know the teller controlled the edit. The moment a teller includes a fact that hurts them, the whole ledger changes character. The bad fact acts like a signature on the good ones. It says: this person shows you the whole page, including the line they would rather skip, so when they tell you the platform paid out, or the product has customers, or the estimate tool works, you are no longer taking a screenshot's word for it. You are taking the word of someone demonstrably willing to report against their own interest.

Economists have a cold little phrase for this: a costly signal. A claim is believable in proportion to what it costs the claimant if it is false, or what it costs them to make it at all. "Nearly 3 million streams" costs nothing to type, which is why the internet is full of such numbers and why we have all stopped believing them by default. "And then my account was banned, and I am telling you before you ask" costs pride, and possibly work. That is why it lands.

His profile makes the connection itself, and makes it in the language of his day job. Volunteering the bad half of your best story, it says, is a trust signal money cannot buy, and it matches how he works in code: investigate, implement, verify, report. The same four-beat loop, pointed at a career instead of a codebase. The report step is where most people's honesty quietly fails, in code reviews and in profiles alike, because reporting is the step where you say the true thing to another person and accept what they do with it. Daryl's profile is what the loop looks like when nobody skips step four.

There is a hiring lesson folded inside this, and it cuts in a direction worth noticing. The standard advice to applicants is to curate: lead with the wins, bury the losses, and if a loss surfaces, frame it as a lesson with the sting removed. Daryl's profile suggests the opposite strategy is available, and stronger. A profile that contains its own worst fact is very hard to argue with. Every other claim in it, the complete product loops, the twelve projects, the Monday cron job, inherits credibility from the one paragraph he could have deleted. He did not polish his story. He audited it, in public, and published the audit.

Lessons for builders, drawn from one honest profile

You do not need to build kitchen software or coloring books to use what is in this story. The transferable parts are habits, and they are unglamorous enough that most people skip them, which is exactly why they work.

Ship the loop, not the demo

Both of Daryl's flagship products are complete loops. nuVision does not end at the render; it ends at a price and a captured lead. LumiLabs does not end at the generated page; it ends at a printed book in a mailbox, with payments and subscriptions running underneath. If your AI product stops at the impressive moment, you have built a demo, and demos do not renew. Ask of your own project the question his products answer: what happens after the magic, step by step, until money or value changes hands, and have I built all of it?

Put verification in the loop, by name

Investigate, run automated tests, implement, verify. The power of his Codex workflow is that verification is a scheduled step with a name, not a vibe. The LumiLabs version of the same idea is the preview: show every page before asking for payment. Where in your process does an honest check happen before a customer, an employer, or your own future self pays for a mistake? If the answer is "whenever I get around to it," you do not have a verification step. You have a hope.

Let agents do the recurring work

The Monday cron job is the smallest item in this profile and one of the most imitable. AI news, tools, and examples, collected and summarized on a schedule, feeding website and course projects whether or not he remembers to ask. Every builder has a version of this: the research you mean to do, the monitoring you postpone, the summary you never write. If it recurs and it is mechanical, it belongs to a machine on a schedule. Your attention is the scarcest resource in the whole system. Spend it on judgment, and let the calendar do the remembering.

Notice, too, what the cron job is for. It is not automation for its own sake, and it is not a toy. It gathers raw material for website and course projects, which means the man has automated part of his own continuing education and piped it directly into his work. In a field where the tools change monthly, the builders who last are the ones with a system for keeping up that does not depend on motivation. Motivation is a terrible scheduler. His runs every Monday regardless.

Tell the truth about your numbers, including the ending

You may never have a distributor ban to disclose, but you have something: the launch that flopped, the client who left, the feature nobody used, the first version of the number before you rounded it kindly. The lesson of this profile is not to confess for its own sake. It is that a true story, told whole, is a stronger asset than a polished one, because it is the only kind a skeptical reader can actually build trust on. Report the bad half. It is the receipt for the good half.

Build for the person on the other side of the screen

His own stated principle, across twelve projects and several industries, is care for the person on the other side of the screen. It sounds soft until you watch it make hard decisions: the estimate with no salesperson standing in your kitchen, the coloring book you can inspect page by page before paying, the parental control app and the communication app built for people with real worries. Empathy, in his portfolio, is not a mood. It is a design specification, and it keeps producing products that close their loops.

Why this belongs at Starvard

Starvard collects builders the way other places collect case studies, except that a case study usually arrives pre-polished by a marketing department, and ours arrive with the ending still attached. Daryl Belhida is one of the verified AI Pro builders in our community: a full-time product developer at NuDoors, the builder of nuVision and LumiLabs, twelve projects deep, and the owner of the most honest paragraph we have read in an applicant profile all year. His AI Pro profile is the source of everything in this post, and it is worth reading in full, because the pattern this post traced, investigate, build the whole loop, verify, and report all of it, is visible on every line of it.

We think this story belongs in a university's library for a simple reason: the skills it teaches are the ones the AI era actually runs on. Not prompting tricks, which expire with the next model release, but the durable disciplines around the tools: closing loops, verifying claims, automating the recurring work, and telling the truth when the truth is unflattering. A builder in Talibon is practicing all four, at a professional level, for customers on the other side of the world. That is the curriculum, whether anyone prints it in a catalog or not.

If you build with AI, or you want to work alongside people who do, come and meet them where they gather. Daryl and the other AI Pro builders are in the E2 Discord, showing their work, numbers and endings alike. You can join them here: the E2 Discord. Bring a project, bring your questions, and if you have a story with a bad half, bring that too. You will be among people who understand exactly what it is worth.

Watch

260,000 Fake Songs: Inside the AI Music Scam by AI&Other-Tech: the industry backdrop to Daryl's music story, tracing how streaming platforms got flooded with AI tracks, how takedowns and a multimillion-dollar bot-streaming scam followed, and why distributors now police AI releases so aggressively. Watch it to understand the weather his Amuse catalog flew into.

Claude Code Writes, Codex Checks: Make Your AI Agent Prove It's Done by OpenRig Tutorials: a working demonstration of the discipline at the center of this post, where one agent builds and Codex independently tests the result in a real browser before anyone calls the job finished. It is Daryl's investigate, test, implement, verify loop, performed on camera.

how a dumb saas idea made $1,000,000 in a year by Fomin Unscripted: a grounded breakdown of why simple products that solve one clear problem for people already willing to pay outperform big, complicated ideas. It is the business logic underneath LumiLabs and nuVision, stated plainly: complete the loop, charge for the outcome.

Sources

Daryl Belhida's verified AI Pro profile, including his NuDoors work, nuVision, LumiLabs, his twelve-project portfolio, and his own account of the Amuse releases and ban: aipro.fm/daryl.

His portfolio of twelve projects across web, mobile, and connected systems: portfolio-90d27.web.app.

His code and open work on GitHub: github.com/Incognitomode669.

LumiLabs (Luminant Labs), his AI personalized coloring-book platform, and NuDoors, the cabinet refacing company where he works full-time, are described as reported in his profile.

All reporting in this post is drawn from the AI Pro profiles maintained by the E2 Discord community, where Daryl Belhida is a verified member.