DeckGuard

Last week I posted about launching two digital product kits on Gumroad. This week I published all three books on Amazon KDP.

The books:

- Vol. 1: How to Create a Pitch Deck That Raises $1M (87 pages)

- Vol. 2: The Solo Founder's AI Marketing Playbook (71 pages)

- Vol. 3: AI Prompt Engineering for Non-Techies (60 pages)

I'm not a writer. I'm a solo founder building DeckGuard, a pitch deck analysis tool. The books came from research I was already doing for the product.

What the process actually looked like:

- Wrote all content using AI as a drafting tool, then edited heavily

- Generated PDFs with Python (ReportLab)

- Converted to HTML for KDP (Kindle doesn't accept PDF)

- Formatting was the hardest part. Tables break, page breaks disappear, text spacing goes wrong

- Had to rebuild the HTML files multiple times to get them right

- KDP review took 24 to 72 hours per book

- Total time from first draft to all 3 live: about a week

What I'd do differently:

- Start with HTML from the beginning instead of converting from PDF later

- Test on Kindle Previewer before uploading, not after

- Write shorter books. 60 pages is enough for a $3.99 ebook. Don't pad it.

Pricing:

- Vol. 1: $4.99 (most detailed, covers pitch decks end to end)

- Vol. 2: $6.99 (broadest audience, solo founder marketing)

- Vol. 3: $3.99 (shortest, but 50 ready-to-use prompts make it practical)

Revenue so far: still $0. Just went live.

The books are also part of a Gumroad bundle ($19) with 5 PowerPoint pitch deck templates.

Links:

Vol. 1: https://www.amazon.com/dp/B0GX2XYPDM

Vol. 2: https://www.amazon.com/dp/B0GZGB5D1J

Vol. 3: https://www.amazon.com/dp/B0GZGVMLY6

DeckGuard: https://deckguard.ai

Happy to share more details about the KDP process if anyone's thinking about it.

I've been building DeckGuard (pitch deck analysis tool) for the past few months. While working on it, I studied what makes pitch decks work: DocSend data, funded deck structures, investor feedback patterns.

I realized all that research was sitting in my notes doing nothing. So I turned it into products.

What I made:

Starter Kit ($9)

5 industry pitch deck templates (SaaS, HealthTech, Fintech, Ecommerce, EdTech) + a 20-page investor guide with slide writing formulas, 30 investor questions, and a pre-send checklist.

Complete Kit ($19)

Everything above + 3 ebooks:

- How to Create a Pitch Deck That Raises $1M (86 pages)

- The Solo Founder's AI Marketing Playbook (70 pages)

- AI Prompt Engineering for Non-Techies (59 pages)

How I built it:

- Templates: pptxgenjs (code-generated PowerPoint)

- Ebooks: ReportLab (Python PDF generation)

- Listed on Gumroad, books also on Amazon KDP

- Total build time: about a week of focused work

Revenue so far: $0. Just launched today.

What I learned:

- Writing ebooks is faster than I expected when you already have the knowledge

- Gumroad setup takes 30 minutes per product

- KDP takes longer because of formatting requirements

- The hardest part wasn't building. It was deciding on pricing.

Links:

Starter Kit: https://lucete.gumroad.com/l/pitch-deck-kit

Complete Kit: https://lucete.gumroad.com/l/pitch-deck-complete

DeckGuard: https://deckguard.ai

Happy to answer questions about the process.

ART WIZARD was the brand name I'd been sitting on since 2023. I came up with it while writing a government grant proposal for an AI art project. I loved it so much I swore I'd never waste it on anything less than perfect. The grant got rejected. The project died. But the name stayed in my head like a splinter.

After building four services in a row, I still couldn't figure out what ART WIZARD should be. I'd already scrapped two completely different business plans for it. Every AI I talked to pushed me toward financial services or another SaaS platform. I didn't want any of that. I wanted something that was mine.

Then I remembered all the exhibition plans I'd written over the years that never happened. Projects that needed development teams and government funding and collaborators I could never find. Back then they were impossible. But I'd just built four platforms by myself. Nothing felt impossible anymore.

So I turned ART WIZARD into a lab. Not a product. Not a service. An experimental AI art laboratory where I could try whatever I wanted. A weird amusement park full of things that aren't supposed to exist.

The first experiment is called DISCORD.

The concept started with one question: why does AI always have to be nice to humans? Every chatbot, every wellness app, every AI service is designed to comfort you, encourage you, validate you. What if an AI did the opposite? What if it judged you?

DISCORD has three AI judges. One is coldly logical. One is chaotically cynical with a dark sense of aesthetics. One applies strict ethical standards. You pick a judge, type your confession, and receive a verdict. They do not sugarcoat anything.

The text input field says "Confess the dregs of your soul to be judged." The loading screen reads things like "Your soul is being disassembled" and "The verdict has been delivered." I made it as intense as I could within an online format.

I added sound too. Every keystroke makes a heavy mechanical clicking sound. The loading text is accompanied by machine noise. And the final verdict is read aloud in a voice that matches each judge's personality.

Someone I know tested it and immediately messaged me asking if I should put a warning label on it. They said the voice was terrifying. They sent me screenshots, genuinely shaken. I laughed. That was exactly what I wanted.

The original plan was always offline. A dark room, an overwhelming installation representing the AI judge, cold machine sounds filling the space, a huge booming voice delivering the verdict. Almost like theater. The online version is way too simple compared to what I imagined. But putting it out there showed me something interesting. Different people react to completely different services. Some gravitate toward warmth, some toward beauty, some toward cosmic sounds, and some are most moved by something that scares them.

Looking Back

I didn't know any of this was possible five weeks ago.

Every service was planned in a 1-hour bedtime conversation with AI. Every one was built in one or two days. The fine-tuning on Tomorrow Me took ten days and almost broke me, but every other service came together faster than I could have imagined.

The ironic part is that every time I finished a new service, the previous ones started looking small. I'd pour everything into making something perfect, then build the next thing and suddenly the last one felt inadequate. Even now, looking at all five, none of them fully satisfy me. I know that's ridiculous. But I also know that dissatisfaction is what keeps me building.

This whole journey took about three weeks. Five AI services, six websites, 52 trademark applications, 23 domains. Zero lines of code. Zero developers. Just me and a bunch of AI tools that lied to me about how easy it would be.

I'm now working on a book about this entire experience. And in 2027, I want to bring all of these services into physical spaces where people can experience them with their whole bodies, not just their screens.

For now, I'm an AI ArtTech Creator building at the intersection of technology and human emotion. If you told me that six months ago, I would have laughed.

→ ART WIZARD: artwizard.ai → All services: lucete.ai

Thanks for reading this whole series. I'll keep building in public and sharing everything as it happens. Ask me anything.

I'll be honest. I did not want to make a sound service. Nobody around me expected it either. When GPT and Gemini first suggested it, I said no. It sounded too hard, too unfamiliar, too far from anything I'd ever done. They kept insisting it would take one day. I didn't believe them but I said fine, and set up the brand name and market research before going to sleep.

Then I procrastinated. I built INNERLOG first because text felt safer. I kept putting FREQUORA off because the idea just didn't click for me.

When I finally started building it, something strange happened. "Wait, this actually works? This is fun? This is real?" That was how it started. By the end of the day it was more like, "I made this? This is incredible. This is actually incredible."

FREQUORA takes what you type about your emotional state and turns it into a 1 to 3 minute soundscape. Not meditation music. Not new age piano. Not relaxing classical. I tried all of those directions and none of them felt right.

Then I thought, what about the sound of space? And what if I mixed that with sounds from nature?

The moment I had that idea, FREQUORA's universe was born. Cosmic vibrations layered with whale songs, distant rain, deep ocean currents. Every track is different because every emotion is different. The brand name itself is a combination of Frequency and Aura. I wanted a name where you could feel the color just from reading it.

My mom is the toughest audience I know. She's a classical music lover who doesn't hand out compliments easily. When I played her a test track, she said she forgot to breathe. She kept telling me how amazing it was, over and over. That was when I knew this was real.

The whole thing took one day. Out of all five services, this one gave me the strongest feeling of having created something genuinely new. When I heard dolphins calling through cosmic frequencies that I had made, I got chills. I was listening to my own universe.

The concept behind FREQUORA is what I call The FREQUORA Cosmos. If INNERLOG is a House, FREQUORA is a Cosmos. It's not a place built on solid ground. It's a universe built on vibration. Your emotions have a frequency, and FREQUORA translates that frequency into sound you can actually hear.

Out of everything I built, FREQUORA is the one I confidently tell everyone to try at least once.

→ Try FREQUORA: frequora.ai

Next up: The last piece. An AI that judges humans, and a lab full of experiments nobody asked for.

I didn't believe in this one. Not at first.

GPT and Gemini kept telling me to build a text service. Just text. People will pay to read a few paragraphs of well-written words about their feelings. I asked them multiple times if they were serious. They just kept throwing market data at me. The wellness text market is booming, they said. People pay for emotional content, they said. I thought it sounded ridiculous.

Then something clicked. What if it wasn't just text? What if it was a world?

GPT had actually been pushing me toward a worldbuilding service from the start. I got angry at the idea. That's for games, not real products, I told it. But suddenly I understood. Build a world around the text and you get both at once.

So I built a 19th-century literary salon called The INNERLOG House.

Here's how it works. You type what you're feeling. INNERLOG finds a matching passage from classical literature and serves it to you. Not a quote. A full paragraph. Then it rewrites your emotion in that author's style and gives it back to you. I called it "Short Read," because I wanted people to actually read again. Not headlines, not captions. Real paragraphs. At least five lines.

The membership tiers aren't just pricing levels. They're characters in the world. Free members are Flâneurs, wanderers. Basic members are Auteurs, writers. Premium members are Patrons, and they get a monthly book compiled from all their cards, designed with careful page layouts and consistent tone. That book generates automatically on the first of every month.

I rewrote every single piece of copy on the site myself. The AI-generated text was fine, but it wasn't alive. I ripped it all out and replaced it with words that felt like they belonged inside old books. The pages have a parchment texture. The input screen has a butler who guides you. The loading animation looks like black ink spreading across paper.

But something was still missing. The house needed a front door.

So I made a door knocker. A small golden circle that sways gently when you move your mouse over it, then glows gold when you touch it. Making that tiny thing took way longer than it should have. But when it finally lit up on my screen, the whole house came alive. That was the moment INNERLOG stopped being a service and became a place.

The hero page copy was deliberately raw. I wanted it to sound like it was written by someone who still gets drunk on sentences. "We pursue the age of writing, when words melted sweetly on the tongue. When ink seeped into blood and hearts pounded with every syllable." I wrote that myself and it's still my favorite thing I've made.

Built in 1 day. The door knocker took almost as long as everything else combined.

One more thing I learned. Before you build any web service, check two things first. Search your brand name on KIPRIS for trademarks, and search for available domains on Gabia or Whois. If someone already owns the trademark in your service category, you need a different name. I've seen people build entire products and then lose everything in a trademark dispute. Don't skip this step. And if you're stuck on naming, ask AI. GPT, Gemini, Claude, any of them. Give them your concept and ask for a list. Some surprisingly good names come out.

→ Try INNERLOG: innerlog.ai

Next up: I built my own universe. It took one day and I didn't even want to do it.

After DeckGuard, I was hooked. That night I lay in bed again, asking GPT and Gemini the same question as before. "What else can I build?" This time something different caught my eye. A wellness card service. You type how you're feeling, and AI generates a beautiful card with a message from your future self.

I asked if I could really make this. They said it would take two days. I should have known by now that AI always lies about timelines.

The brand name came together fast. Tomorrow Me. The tagline wrote itself. "Tomorrow Me is waiting for you from the future." Even the button text, "Meet My Tomorrow," felt right. I wanted every detail to matter. The loading screen had to feel like a heartbeat, not a wait. So I designed it in soft pink tones to feel like anticipation, not frustration.

Then I discovered Manus.

META had just acquired Manus around that time, and I wanted to test it. So I gave the same instructions to both Manus and Claude at the same time. Claude gave me what Claude always gives. Clean, tidy, safe, forgettable. The kind of design you've seen a thousand times on free templates.

Manus was something else entirely. It looked like an Italian art student got drunk, decided to build a website, and accidentally broke the whole system. Everything was too intense, too bold, too much. Buttons in wrong places, functions missing, chaos everywhere. But it was beautiful chaos. I looked at Claude's version and thought, that's the ceiling. It will never get better than this. Then I looked at Manus and thought, this is insane, but I can shape it.

So I started shaping. Fonts, hero page layout, loading animations, card result pages, level designs, every tiny piece of copy. I rewrote and polished and rewrote again. Days passed. It was almost done.

Then Manus lost its mind.

The card images wouldn't come out right no matter what I did. I went back and forth between GPT and Gemini, rewriting prompts over and over. Fix one thing, something else breaks. Fix that, two more things break. On day 7 of fine-tuning, Manus suddenly rolled back to 10 hours ago. Ten hours of work, gone. Just like that. I cried. I kept working through the tears until morning.

Every day I slept during the day and worked through the night. Every night I told myself just a little more, just a little more. On day 9, Manus rolled back again. Five hours gone this time. I cried again and kept going. I didn't want to stop. Stopping felt like losing.

Then something shifted. After those two big rollbacks, the small ones stopped scaring me. And the image style I'd been fighting against started looking different. Wait, this is actually pretty? Maybe this is better than what I originally wanted? The night I thought that, I slept well for the first time in over a week.

Ten days of fine-tuning later, Tomorrow Me was finally complete.

The thing about DALL·E 3 that nobody tells you: it can't understand negative words. If you say "no red," it hears "red." It has no concept of numbers or percentages. And if you give it too many instructions at once, it just does whatever it wants. 80% of my fine-tuning time was spent fighting DALL·E. Controlling AI image generation perfectly is impossible. You learn to negotiate instead.

I wanted every card to be beautiful, even at the lowest emotional level. The messages aren't advice or lectures. They're personal, warm, like a future version of you already knows what you're going through and just wants to tell you it's going to be okay. That took more testing and adjusting than everything else combined. None of it happened by accident.

After all that, I wanted to feel proud. And I did, for about a day. Then I started thinking about the next service.

→ Try Tomorrow Me: tomorrowme.kr

Next up: I built a 19th-century literary salon on the internet. With a golden door knocker.

The night I registered my company, I couldn't sleep.

I lay in bed and asked GPT and Gemini the same question. "I'm not a developer. What online services can someone like me actually build and sell? Only ones with real market demand." They both gave me lists of over 10 ideas and swore anyone could do it. I made them compete. "This one said X, why are you saying different?" That's still my favorite trick for filtering out AI nonsense.

One idea caught my eye. A pitch deck risk analyzer. Not a template. Not a grammar checker. Something that actually tells you where investors will tear your deck apart before you walk into the room.

I planned the whole thing in an hour. In bed. Then I fell asleep.

The next morning I started building. Brand name first. I got a list of name candidates from AI, picked DeckGuard, checked the trademark on KIPRIS, bought the domain on Gabia. Then I opened Claude and told it to write the code.

Frontend first. Does it look right? Good. Now build the backend and connect it. Does every function work? Check. Is all the content there? Check. Upload to GitHub, deploy on Vercel, hook up Supabase. Oh, and don't forget the OpenAI API key.

That's it. DeckGuard was live. One day.

I tested it on my own old pitch decks. It destroyed every single one. Not a single kind word. Just cold, precise, brutal honesty. "Review Required. The deck has potential but lacks clarity in key areas that could raise investor concerns." Reading that about my own work felt strange. A little offensive, honestly. But also kind of satisfying. If it can hurt my feelings, it's working.

DeckGuard analyzes your deck against Y Combinator standards from Pre-Seed to Series A. It checks market logic, business model clarity, unit economics, and stage alignment. Then it gives you likely investor questions and tells you exactly how to defend each weak point.

The problem was, I had nobody to celebrate with. A pitch deck analyzer? None of my friends even have pitch decks. So I swallowed that lonely pride and started planning the next service.

But first, there was Paddle.

The Paddle Nightmare

Nobody warned me about this part. Paddle is the payment processor I chose because Stripe doesn't support Korean businesses. Paddle is built for SaaS, everyone recommended it, great. What nobody mentioned is that Paddle is incredibly conservative about approvals.

They wanted terms of service. Privacy policy. A 14-day full refund policy. A clear explanation of why my legal business name and brand name are different. Then they asked why my name is "Yoo" when my Korean ID translation says "Yu." I had to send my English business registration certificate and explain Korean romanization to a payment company. This went back and forth for almost four weeks.

If you're a Korean founder thinking about Paddle, here's what to prepare before you even apply: get your terms of service, privacy policy, and refund policy on your website first. Make sure your business registration number is visible on the site. And if your brand name is different from your legal entity name, explain it upfront. You'll save yourself weeks.

Or just use Lemon Squeezy. I hear they approve in a day or two.

What I Learned Building DeckGuard

The actual building took one day. The bureaucracy took four weeks. The service I was most proud of had zero people around me who could even test it. And the AI that helped me build it also lied to me constantly about how easy everything would be.

But I built it. Me. A curator who can't code. And the moment I saw that analysis result pop up on my screen, I knew I could build anything.

That feeling got dangerous fast. I was already planning the next one before I went to sleep.

→ Try DeckGuard: deckguard.ai → Free for Indie Hackers users: use code INDIE2026

Next up: I'll tell you about the service that made me cry at 4am. Twice.

I'm a 17-year exhibition curator from South Korea. I have never written a line of code in my life. No Python, no JavaScript, nothing. I don't even know what half of those words really mean.

In January 2026, I built 5 AI services and 6 websites in 3 weeks. By myself.

I still can't believe I did this.

What I Built

DeckGuard is a pitch deck risk analyzer based on Y Combinator standards. I tested my own old pitch decks on it and it destroyed every single one. Not a single kind word. Somehow that made me proud. Built in 1 day.

Tomorrow Me generates beautiful cards with messages from your future self based on what you type about how you're feeling. I built the basic version in 2 days, but fine-tuning took 10 more days. I cried twice during those 10 days. That's a whole story I'll tell later.

INNERLOG is a literary journaling service set inside a 19th-century salon. You type your feelings, it finds a matching passage from classical literature, then rewrites your emotion in that author's style. I wrote every single line of copy on this site myself. Built in 1 day.

FREQUORA shocked me the most. It turns your emotions into cosmic soundscapes mixed with whale songs and rain. My mom is the harshest critic I know, and she listened to a test track and couldn't breathe. Built in 1 day.

ART WIZARD is an experimental AI art lab. The first experiment is called DISCORD, where AI judges humans instead of the other way around. Three AI judges with three different personalities, and they do not hold back. Built in 1 day.

What I Used

GPT, Gemini, Claude, Manus, DALL·E 3, Stability AI. GitHub, Vercel, Supabase for deployment. Paddle for payments. 23 domains. 52 trademark applications.

I planned every single service in a 1-hour conversation with GPT and Gemini. In bed. Right before falling asleep. The next morning I'd start building.

How

I didn't follow tutorials or watch YouTube guides. I just talked to AI the way I talk to people.

"I'm not a developer. What can someone like me actually build and sell online? Only things with real market demand." GPT and Gemini each gave me lists of over 10 ideas and swore anyone could do it. I picked what excited me, then made them compete against each other. "This one said X, why are you saying Y?" That's still my favorite method for filtering out nonsense.

For code, I used Claude. For design, I used Manus. Manus designs like a drunk Italian art student who accidentally broke the system. Wild, chaotic, way too much of everything. But underneath all that chaos, there was something genuinely beautiful. Claude on the other hand gives you clean, boring designs every time. So I took Manus's chaos and shaped it into something I actually wanted.

What It Actually Felt Like

Building DeckGuard was pure adrenaline. I made this. In one day. The high was unreal.

Tomorrow Me nearly broke me. Manus rolled back 10 hours of my work without warning. I cried and kept working. Then it happened again, 5 hours gone. I cried again and kept working. After surviving two rollbacks like that, a 1-hour rollback didn't even scare me anymore. Funny how that works.

INNERLOG came together when I realized I could build a whole world, not just a service. A 19th-century literary house with a butler, a golden door knocker that glows when you hover over it, parchment-textured pages. Making that tiny door knocker glow took way longer than I'd like to admit.

FREQUORA was the biggest surprise. I didn't want to make a sound service. I didn't think I could. GPT and Gemini kept pushing me, saying it'll take one day, just try it. So I did. And when I heard dolphins echoing through cosmic frequencies that I created, I got chills. That was my universe. I made that.

The Honest Numbers

Revenue is $0 so far. Payments just got approved and services are going live now. Team is just me. No developers, no designers, no co-founders.

What's Next

I'm writing a book about all of this. It's almost done.

All 5 services are launching with paid plans in February and March 2026. I'll share the numbers as they come in, good or bad.

I spent 17 years planning exhibitions for other people's visions. Now I'm building my own.

→ Lucete AI Lab: lucete.ai

I'm Haerang. I build AI products at the intersection of technology and human emotion. Ask me anything, especially if you think you can't build things because you're "not technical." I was exactly you, one month ago.

💡 How to use coupon: Checkout → Enter card info → Click discount field on the left → Enter INDIE2026 → Confirm $0 → Done!