MeerMost

The cobbler finally got his shoes: how I built my portfolio site after 3 years of client work

For three years I built websites, landing pages, and apps for clients. Work came in through word of mouth, I wasn't losing jobs for lack of a storefront, and I kept putting it off. In March 2026 I finally sat down and built meermost.site. This is an inside breakdown: what I chose, why exactly that way, and where I'd have saved myself a week if I were starting over.

Why there was no website for three years

Honest answer: I didn't need one. Clients came through acquaintances, developer chats, and my Telegram channel. Every new order was the result of a specific conversation, not "came from the website."

What changed:

  • There were more orders than time. The manual "I'll explain it on a call" mode stopped being enough. I needed a place to send people before the call so they could rule out what wasn't relevant on their own.
  • My own products grew to a size that doesn't fit in a single message. MeerGuard, MeerBot, Meerno, plus a dozen client case studies. I needed a place where it all sits as a portfolio.
  • A video production team appeared. Video services need to be shown with examples, not descriptions.
  • YouTube became the main acquisition channel. Videos lead to the site, and the site has to convert into a request.

Putting it off further became more expensive than the decision to build it.

Stack

Next.js 15 (App Router) + React 18 + TypeScript. I was tempted to go with Tilda or Framer and put it together over a weekend. I chose code. Three reasons:

  1. Control over SEO. Canonicals, hreflang, structured data, dynamic sitemaps and robots — on website builders these are either missing or limited to paid tiers.
  2. Calculators. I knew the main SEO pages would be the app cost calculator and the video calculator. You can't build a step-by-step wizard with a client-side formula on Tilda.
  3. It's part of the portfolio. The site itself shows what I can build. A website builder would have broken that.

Tailwind CSS. No deliberation here. For a one-page portfolio with fifteen pages and forty components, it just works. No CSS-in-JS, no separate style files.

Vercel for deployment. The free Hobby tier covers everything I need. Previews on every commit, edge functions, analytics, cron jobs. A separate VPS would cost me half a day a month on updates and monitoring; here it just works.

Content architecture

The site consists of five types of entities:

EntityWhere it livesFormat
Translationslib/i18n/en.ts, lib/i18n/ru.tsTypeScript objects with a source type in English
Projectslib/projects.tsTypeScript array with metadata
Case studieslib/case-studies.tsTypeScript objects with bilingual content
Blogcontent/blog/<slug>.<locale>.mdMarkdown with frontmatter
SEO schemaslib/seo/schema.ts, lib/seo/metadata.tsTypeScript builders

The key decision here is bilingualism through types. The English version of the translations is the source of truth, and the Translations type is inferred from it. The Russian version is typed as Translations. If I add a key to English, TypeScript complains about Russian until I fill it in. That's better than any YAML or JSON translation system. You physically can't forget a translation — the project won't build.

i18n without a library

I didn't use next-intl, next-translate, or similar libraries. Instead, a simple structure:

  • app/[locale]/.... Every page sits under the locale parameter.
  • lib/i18n/index.ts. locales = ['en', 'ru'] as const, defaultLocale = 'ru'.
  • Components take locale and t (translations) as props.

Why this way:

  • Fewer dependencies. Any i18n library brings its own approach to routing, namespaces, and pluralization. For two languages that's overkill.
  • Full control over SEO. hreflang, canonical, localized URLs are written by hand in one file, lib/seo/metadata.ts. I know what's rendered in <head> on every page.
  • Server components by default. A translation is a plain object, not a React context. On the server it renders to HTML without hydration.

One downside. There's no automatic language detection via Accept-Language. I closed that gap manually with middleware.ts redirecting to defaultLocale.

Blog on markdown

The blog is built on static markdown files with frontmatter:

content/blog/cobbler-finally-got-shoes.ru.md
content/blog/cobbler-finally-got-shoes.en.md

One file, one article in one language. Parsing via gray-matter, rendering via react-markdown + remark-gfm. lib/blog/articles.ts holds the file-reading logic, frontmatter validation, sorting by date, and finding related articles by tags.

Frontmatter:

title: "..."
description: "..."
date: "2026-06-10"
category: development | ai | business | marketing
tags: [...]
videoId: "..."   # optional, for embedding YouTube
draft: false

Why markdown and not a CMS like Notion / Strapi / Contentful:

  • Git as version control. Every change is a commit. You can roll back, look at the history, make a branch for a draft.
  • Local editing. I write in a regular code editor, without switching to a browser. That matters for long texts.
  • Simpler deploys. Markdown is part of the repo. Changed it, pushed, Vercel rebuilt. No webhooks from a CMS, no delays.
  • I don't pay for a tool. Notion-as-CMS means a subscription plus API request limits.

The blog's video format deserves a separate note. If there's a videoId in the frontmatter, the article renders with a YouTube embed at the top and a transcript at the bottom. That's needed for articles I make from my own YouTube videos. The video itself meets the need, while the text works for SEO and for those who read faster than they watch.

Calculators as the Main SEO Pages

The app cost calculator and the video calculator aren't site features — they're landing pages. I plan to drive the bulk of my Google and Yandex traffic to them on queries like "how much does an app cost" and "how much does a video for business cost."

Architecturally, they're step-by-step wizards. Five steps, the formula runs on the client, no backend for calculations. The ruble exchange rate is hardcoded, because for an app estimate a 2–3% margin of error doesn't matter, and the overhead of calling an exchange rate API is just another point of failure.

The main find. Automatically showing the final screen sends an anonymous lead to Telegram. Without waiting for the user to fill out the form. If someone makes it to the end of the calculator, I know about it right away and can prepare for the conversation in advance.

Deduplication via sessionStorage so the same user doesn't generate ten identical notifications per session. Protection against instant recalculations is a simple debounce.

SEO Infrastructure

In lib/seo/:

  • metadata.ts. pageMetadata() takes a page and a locale, returns a Next.js Metadata object with canonical, hreflang, OG images, Twitter Card.
  • schema.ts. JSON-LD builders for Organization, Person, BlogPosting, Service, Product.

In app/:

  • sitemap.ts. Dynamic sitemap.xml: pages, projects, blog articles. Bilingual URLs automatically.
  • robots.ts. Dynamic robots.txt with the current domain.

No manual sitemap edits when you add an article or a project. Add an .md file and it automatically lands in the sitemap. Add a project to lib/projects.ts, same thing. This eliminates a whole class of errors that plague sites built on website builders.

The domain is meermost.site. Originally it was meermost.vercel.app. I bought .site instead of .com because a free .com with my surname wasn't available, and .site works for SEO exactly the same way.

Telegram Integration

All forms and leads go to a single Telegram chat via lib/telegram/send.ts. Six sources:

SourceTrigger
Contact formSubmit on /contact
App calculatorFinal screen of the calculator shown
Video calculatorFinal screen of VideoCalculator shown
Vercel deployWebhook from Vercel on deployment.succeeded / deployment.error
CapCut cronThe 17th of every month at 12:00 MSK, reminder to pay for the subscription
MeerBot demoVia the demo form calculator

All analytics and all day-to-day contact flow into one place. For solo mode, that beats splitting across email, CRM, and Slack. Zero context switching.

What I'd Do Differently

Start with landing pages built for traffic instead of the portfolio. I started with the homepage, then about me, then projects. I built the calculators and the video landing page last. It would have made more sense to start with them — they bring in leads, while the portfolio only confirms competence.

Two languages from the start. I did Russian first, then added English. In the end I had to rework half the components so they'd accept locale and t. If I'd built i18n in from day one, it would have taken less time.

Not trying to do everything in one pass. I built the portfolio, services, blog, video production, and calculators in one long sprint. It would have been better to split it into two waves: first portfolio + contacts + app calculator, then everything else. That way I wouldn't have burned two weeks and would've been getting feedback along the way.

A content plan before the site, not after. Right now I'm writing articles and service pages after the fact. If I'd done the content plan first, the site structure would have been built around it instead of around my fantasy of how a studio should be organized.

What My Own Site Taught Me

Your own product is always harder than a client's. When you're building for a client, there's a brief, a deadline, someone else's opinion, a line you can't cross. When you're building for yourself, every line can be moved. And you move them, week after week, until you run into reality.

The solution turned out to be simple. Set a limit on myself and lock in specific items. The result is a site with a clear structure that works to bring in leads instead of sitting there as "a portfolio for show."

What's Next

Want your own site or landing page that works to bring in leads instead of existing just to exist? Estimate your budget or get in touch — let's talk specifics.

All articles

Related articles

AI Search Promotion: How a Site Made It Into ChatGPT Answers
Marketing

AI Search Promotion: How a Site Made It Into ChatGPT Answers

GEO without promises: two leads from ChatGPT in October 2026, the tag that makes them visible, what was already in place on the site, and what can't be guaranteed in AI search promotion.

8 min read
SEO Articles: How to Publish a Steady Stream Without Getting Hit by a Filter
Marketing

SEO Articles: How to Publish a Steady Stream Without Getting Hit by a Filter

How an SEO article pipeline works: where to source topics, how to pick clusters, why the ceiling is 8–12 articles a month, and why a human reads every one before it goes live.

8 min read
AI in a Mobile App: How to Integrate ChatGPT
Development

AI in a Mobile App: How to Integrate ChatGPT

AI in a mobile app: the API key stays on the server, per-user limits, caching, moderation and a token-based cost estimate. Meerno as the example.

8 min read