So when I decided to build my next product, I didn’t have to think about which platform. I picked the one I know best, not as a creator (I'll change it), but as someone who has spent a decade on the other side of the screen.

Here’s what pushed me over the edge. A close friend of mine is a Catholic priest with a YouTube channel of over 2 million subscribers. I help him with the tech side. His small team (me included, on my free time) manages the channel, reads comments, and tries to understand what resonates. But when you have thousands of comments across hundreds of videos, reading them manually doesn’t scale. The signal gets buried in noise. The team wanted better visibility into what the audience is actually saying, what they’re asking for, what they love, and what frustrates them.

That problem stuck with me. And I realized it’s not unique to big channels. Any creator with a growing audience faces the same thing: the comments hold the answers, but nobody has time to mine them.

What Perspion does

Perspion lets you point at any YouTube channel, select the videos you care about, and run an AI-powered analysis on every comment. A couple of minutes later, you get a full breakdown:

• Content Radar: What topics your audience is requesting, ranked by how often they come up. You never have to wonder “What should my next video be about?”

• Sentiment Map: How people feel about each video. Not just “positive vs negative” but what specifically they loved and what needs attention.

• Question Bank: Every question from your comments, extracted and grouped. Thirty-four people asking the same thing? Now you know.

• Actionable Insights: Plain-language recommendations based on what the data shows.

The name comes from Latin. "Perspicere" means to see clearly, to perceive, to look through. It probably doesn't sound like a typical SaaS name, but then again, a fruit company named itself Apple and did fine, actually way better than fine. My previous product, Gradon, comes from "gradus" (step, progress). Latin roots are a WheyDev thing now.

How I built it: The 17-prompt framework

Before writing any code, I wrote 17 architecture prompts. Each one is a complete brief for Claude Code covering one step of the build: exact database schemas, UI layouts, API structures, copy, error handling, everything.

I spent more time writing the prompts than I’ll spend executing them. The prompts are the architecture. The code follows.

Here’s the sequence:

Foundation (Prompts 1-2)

Project setup and internationalization. English and Portuguese from day one (at least for Perspion). Exact fonts, colors, and folder structure. Setting up i18n early means every component built after this is automatically bilingual. Retrofitting it later is painful.

Surface (Prompt 3)

The landing page. Dark mode, premium aesthetic, seven sections with specific copy and layout. This page can go live before a single feature works.

Plumbing (Prompts 4-6)

Google authentication, the database schema with row-level security so users only see their own data, a YouTube channel with video search, and the comment fetching pipeline.

Core value (Prompts 7-8)

The AI analysis engine that categorizes every comment (content request, question, positive feedback, negative feedback, discussion, spam) and the results dashboard with seven visual cards. I believe this is why people pay.

Retention (Prompts 9-10)

Analysis history with permanent links, search, and pagination. Usage limits and billing tiers (Free, Pro, Agency) with proper enforcement. This turns a tool into a business.

Premium (Prompt 11)

PDF export for paying users. Full formatted report, downloadable and shareable.

Professional (Prompts 12-15)

Error handling for every failure mode, a public changelog, comprehensive project documentation, terms of service, and privacy policy covering everything from subscription cancellation to service discontinuation, written for a real business entity.

Measurement (Prompt 16)

Analytics with a defined conversion funnel: channel searched, video selected, analysis started, analysis completed. Plus monetization signals: When someone hits their free limit, do they click "upgrade"? You can't improve what you don't measure, but tracking without a plan just gives you noise.

Security (Prompt 17)

Two-layered audits before launch. A general sweep covering secrets, unprotected routes, dependency vulnerabilities, and missing security headers. Then, a database-specific deep dive checking that access policies work correctly and simulating what an unauthorized user could see.

The reusable sequence

Strip away the YouTube-specific parts, and you get a framework for most of any SaaS MVP:

1. Stack setup

2. Internationalization

3. Landing page

4. Auth + data model

5. Core input UX

6. Data pipeline

7. Processing logic

8. Results display

9. History + persistence

10. Billing + limits

11. Premium features

12. Polish + errors

13. Changelog

14. Documentation

15. Legal

16. Analytics

17. Security audit

I’m planning to use this exact sequence for every WheyDev product going forward. The prompts change, but the structure doesn’t. At least until I find a better solution. I'm always improving as a professional.

And honestly, 17 prompts were just the starting point. After running them, I created additional prompts for adjustments, refinements, and things I only noticed once I saw the product running. That’s normal. As a friend of mine says, software is a living thing. It always evolves. The framework gives you the skeleton. The flesh grows over time.

Try it

Perspion is live right now at perspion.whey.dev. Three free analyses and no credit card required. Point it at your channel and see what your audience has been trying to tell you.

This is a side project I'm building in my spare time, and I'll keep sharing how it goes here and on social media.

Perspion is the second product from WheyDev, a micro-SaaS studio building tools for creators. The first, Gradon, is a career progress tracker at gradon.whey.dev.

Last weekend, that obsession became a product.

I've always cared about more than the code. Data, analytics, user behavior, product decisions. Those were never "not my job." I just never made it official until now.

It made me ask a real question: am I actually capable of building a product end-to-end? Not just the UI. Everything. Design, frontend, backend, database, auth, payments, AI integration, security, performance.

I needed to find out.

So I built Gradon over a weekend.

With Claude Code, from scratch: a minimalist frontend, database, background jobs, API routes, domain and subdomain setup, Gemini integration, and payments. Something I had never done before.

It’s a career tracking app. Because no existing tool ever matched how I actually think about documenting work. So I stopped waiting for someone else to build it.

One intentional decision: I didn’t integrate GitHub yet. Not because I can’t. Because I wanted to see if someone would track their work even without automation. Even if it’s a quick, manual process. Turns out, they will. That tells me something.

Is it ready to sell? No! I’m telling you straight: don’t pay for it right now.

I built this for myself first. To prove something to myself. And it worked.

But day zero already taught me something I didn’t expect.

I shared Gradon in the Small Bets community looking for early feedback. Someone visited the landing page and didn't sign up because they couldn't picture what the product looked like inside. That's not a complaint. That's a product insight. And within 24 hours I had a roadmap change.

I’ve had a changelog page since day one. But this told me I need more: real onboarding, guided flows, zero friction, always a support link visible. User first, from the very first click.

Here’s what’s coming next:

• A video on the homepage showing exactly how the product works, updated every time something changes.

• Weekly and monthly AI-powered insights with PDF export.

• Anonymous peer comparison, so you can benchmark your growth against other engineers.

• Deeper career analysis over time.

That’s the Product Engineer mindset I’ve been building toward. And now I have a real product to practice it on.

This is WheyDev. My space to build real software, ship real products in my spare time, and become a better engineer in the process.

Gradon is the first one.

If you want to follow someone building in public, stick around.

gradon.whey.dev

--

Keep committing,
Eder

I opened a doc. Copied. Pasted into 15Five. 15 minutes, done.

Not because I’m more organized. I mean, I am. But that’s not the point. I just stopped trusting my memory years ago.

I've been doing this since 2021. Every year since then, promoted. Every single one. Not saying this guarantees anything. But something’s working. I did it wrong at first, though. I only tracked numbers. PRs merged, tickets closed, bugs fixed. Thought volume meant value. It doesn’t. Managers don’t care that you closed 42 tickets. They care about money. Did you make money or save money? That’s the whole game. Everything else is noise. And I agree with that.

So now my Notion doc looks like this:

• TL;DR at the top. Semester highlights, impact only.

• Then Projects split into Features and Fixes. Date, what changed, link to PR.

• I also track Courses, Books, and stuff I did outside work, like blog posts.

• And Investments in my own growth (actual dollars, because I take this seriously).

10 minutes a week. Feels pointless in the moment. Then January comes and everyone’s panicking, and you’re just... done.

The real shift was how I write the summaries.

I used to write “shipped invoicing feature.” Now I write “shipped invoicing automation that cut 3 hours of manual work per week from ops.”

First one is a task. Second one is dollars. Dollars get remembered in budget meetings.

If you can’t connect what you did to revenue or savings, keep asking why it mattered until you hit money. Usually takes two or three whys

In 2023, I was diagnosed with Asperger’s. Lot of things made sense after that.

I like systems. Like routine. Like doing the same thing every week the same way. Most people find that tedious. My brain finds it necessary.

I used to think this was a limitation. Now I see it differently. The thing that makes me “too rigid” is the same thing that means I never show up to a review empty-handed.

Three phases so far:

1. 2021 to 2024, manual. One doc per semester, wrote everything by hand.

2. 2024 to 2026, Google Sheets. Added charts, consolidated years.

3. Now, 2026 forward, AI-assisted. Exploring agents that document through conversation. Goal is to make this invisible.

Also started pulling tasks from company's OKRs directly. So I’m not just doing work. I’m doing work leadership already said matters. Then documenting that I did it.

10 minutes a week sounds like overhead.

Then January comes. Everyone’s reconstructing their year from memory. You open a doc, and you’re done.

Some people call that discipline. It’s not. It’s just writing things down because you won’t remember. Trust me.

--
Keep committing,
Eder