I Turned YouTube Into Our Video CMS. It Now Runs for RM0/Month.

I built a full-scale YouTube-native Video CMS that now manages 33,000+ videos, handles 2 million+ daily video views, and costs RM 0/month to run.
Yes, zero.

I built it inside FMT's infrastructure from scratch, over four weeks, using the same stack we already trusted for our main site — Next.js, MongoDB, Cloudflare, and a lot of caffeine.

This isn't a startup pitch or a product launch.
It's just me explaining how I looked at a RM500K/year problem, shrugged, and thought:

"Wait… haven't we already built the world's best video CDN? It's called YouTube."

The RM500K Problem

When you run a newsroom that publishes 30–45 videos a day, you quickly realize that "enterprise" video CMS vendors don't sell software — they sell anxiety insurance.

Every pitch starts the same way: "You need enterprise-grade video delivery with real-time analytics." Then you get the quote and realize that "enterprise-grade" actually means "your finance team will hate you."

The fun part?
Out of their 100+ features, you'll use about 15. The rest exist so they can justify the invoice.

That's when I realized:
YouTube already solves the hard parts.
Unlimited storage, transcoding, analytics, CDN, global playback.
All I needed was the CMS layer.

The Features Nobody Asked For (But Vendors Make You Pay For)

Every enterprise CMS bundle looks impressive until you start unchecking the boxes you'll never touch.

What we actually needed:

• A reliable dashboard that shows real-time stats
• Real-time video sync when content publishes
• Regional content intelligence for Southeast Asia
• SEO and structured data baked in
• Admin tools for emergencies

All of which YouTube's infrastructure provides — just not through a pretty admin interface.
So I built the interface.

The Realization: YouTube as the Database

Most people treat YouTube as an output channel. I treated it as the database.

That one decision flipped everything.

• YouTube API = Source of truth
• MongoDB = Cached mirror for fast reads
• Next.js ISR = Dynamic freshness with static speed
• Cloudflare = Global edge delivery
• WebSub (PubSubHubbub) = Real-time updates

Result: new videos appear on our site within 30 seconds of being published on YouTube.
No polling. No quota spam. No cron jobs waking up every five minutes.

The Architecture: Elegant, Fast, and (Almost) Boring

I didn't reinvent the wheel. I just made it spin smoother.

YouTube Data API → MongoDB (metadata cache)
        ↓
Next.js ISR → Cloudflare CDN
        ↓
   2M+ daily video views

Simple architecture scales better than clever architecture.

I chose Next.js ISR because it's the sweet spot between static and dynamic — it lets pages regenerate in the background while serving cached versions instantly. Even during viral spikes, users don't wait.

MongoDB made sense because our schema evolves daily — playlists, metadata, tags, experiment fields. SQL would've drowned in migrations.

And then came WebSub — the real hero of this story.

The 3 AM WebSub Discovery

I spent three weeks building a polling system that checked for updates every 5 minutes. It worked beautifully — until one Friday, when breaking news hit, we uploaded 60 videos in a day… and exhausted the YouTube API quota before lunch.

Cue the 3 AM debug session.
I was staring at YouTube's docs, fifth reread, when one sentence clicked:

"YouTube supports PubSubHubbub for push-based notifications."

Wait.
YouTube would tell me when videos update?

I implemented it over a weekend. The next Monday:

• Quota dropped 80%
• Update latency fell to 30 seconds
• And I slept like a human again

That was the moment I realized — engineering isn't about writing code; it's about deleting cron jobs.

The Dashboard: More Than Pretty Charts

This is where it gets fun.
The dashboard isn't just an admin panel; it's a mission control center for video operations.

• Total videos: 33K+
• Videos this week: publishing velocity
• Active playlists: health monitoring
• Cache hit rate: 97%

When management asks, "How many videos did we publish last week?" — I don't query databases anymore. I just point at the graph.

Light and Dark (Because News Never Sleeps)

Dark theme for the 2 AM breaking-news team.

Light theme for 10 AM editorial reviews.

I added both because newsrooms don't run on business hours — they run on adrenaline, caffeine, and push notifications.

Content Intelligence: When the CMS Starts Giving Ideas

But a pretty dashboard is useless if it can't help you make decisions.

That's why I built Content Intelligence — a module that taps into Google Trends across Malaysia, Singapore, Indonesia, and Thailand.

It flags trending topics that we haven't covered yet and suggests video ideas with "opportunity scores."

Example:

• "Johor Water Crisis" trending → alert pops up → editorial jumps in
• The next day, the video hits top 5 across channels

The system learned to act like a junior producer with zero ego and 100% uptime.

SEO & Structured Data: Built Right In

This CMS wasn't just about publishing videos. It had to make them discoverable.

Every video page automatically generates:

• Open Graph metadata for social sharing
• YouTube-linked schema.org JSON-LD for rich snippets
• Canonical and alternate URLs for proper indexing
• Auto-revalidating sitemaps that update in real-time

Google loves consistency. This setup gives every video equal SEO weight as a native article — and it works.

FMT's video pages now appear in Google Discover, Top Stories, and YouTube SERP carousels simultaneously. That's a trifecta most publishers pay extra for.

Playlist Management That Actually Works

Playlists are the backbone of our YouTube structure. But YouTube's own interface makes managing 100+ of them painful.

My CMS turned that chaos into order:

• Real-time playlist sync from YouTube
• Automatic drift detection (catches when counts mismatch)
• Weekly verification before quota reset
• Human-readable analytics per playlist

Before this, a single playlist audit could take hours. Now it's 10 minutes and one button.

The Admin "Oh-Shit" Tools: Production-Grade Crisis Management

I learned the hard way that real systems break in the least convenient ways possible.

1. The Cache Disaster

One night, a deleted video kept showing up on the site. CDN cache hadn't purged. Users complained. Editors panicked.

The problem? Multi-layer caching:

• Cloudflare edge cache (15 min TTL)
• ISR page cache (1 hour revalidation)
• LRU memory cache (5 min)
• MongoDB still had the record

Clearing one layer wasn't enough. The video kept reappearing like a bad penny.

Next morning: I built a "Clear Everything" button that purges Cloudflare, ISR, MongoDB, and in-memory caches in one go. It blocks until every layer confirms purge.

Now it's my most-used feature — and my favorite insurance policy.

2. The Purge Protocol

9:15 PM on a Tuesday. Legal team Slack message:

"That video from this morning — we need it gone. NOW."

The video had already been cached across Cloudflare's edge network, sitting in ISR pages, embedded in sitemaps, indexed by search engines. Deleting from YouTube wasn't enough.

I built the Purge tool that night. One click initiates:

1. Remove from MongoDB (all collections)
2. Remove from playlist associations
3. Purge Cloudflare cache (via API)
4. Clear ISR cache for video page
5. Revalidate homepage and playlists
6. Update sitemap
7. Log everything with admin name, timestamp, reason

Cannot be undone. By design — forces careful consideration before clicking.

Used it three times in the first month. Compliance without chaos.

3. The Sync Emergency

When automation fails (and it does), there's a manual "Pull All Videos" command.

It processes all 100+ playlists systematically:

• Shows real-time progress
• Deduplication protection (can't run twice simultaneously)
• Detailed logging (playlists synced, videos added/updated, errors)
• Takes 10 minutes for full sync

Because sometimes, you need to hit refresh on life.

Each of these tools was built out of pain — and used out of peace of mind.

How It Handles 2M+ Daily Views

The system survives traffic storms because it respects one rule:

"Every request should hit cache."

The 5-Layer Cache Strategy

• Layer 1: Browser cache (5 min)
• Layer 2: Cloudflare edge cache (15 min)
• Layer 3: Next.js ISR static regeneration (1 hour)
• Layer 4: LRU in-memory cache (5 min)
• Layer 5: MongoDB fallback (only on cache miss)

Result: 97% cache hit rate.
MongoDB barely breaks a sweat.

When a video goes viral (like our Anwar or Messi clips), Cloudflare absorbs the hit. Our origin servers don't even flinch.

The Features That Matter (And the Ones We Skipped)

You don't need "AI-powered dashboards" or "video engagement funnels."
You need fewer moving parts and one system you can trust to stay up when the story breaks.

Lessons I'd Tell My Past Self

Read the Docs (Then Read Them Again)

WebSub wasn't magic; it was buried in a sentence I skimmed three times.
The fifth time I read it, I found a weekend's worth of work that replaced three weeks of code.

Lesson: Documentation fatigue is real, but so is missing the one feature that changes everything.

Build Tools Before Features

I built the "Oh-Shit" buttons after disasters struck. If I had done it earlier, I'd have saved many heartbeats and coffees.

When you're building production systems, operational tooling isn't a "nice to have" — it's the difference between sleeping and staring at Slack at 3 AM.

Profile, Don't Guess

I once spent two weeks optimizing thumbnails. Pre-fetch, lazy load, WebP conversion, blur placeholders — the works. Gained 0.3 seconds.

Then I looked at MongoDB queries. No indexes on frequently queried fields. Added indexes in 30 minutes. Gained 2.5 seconds.

Lesson: Your intuition about bottlenecks is probably wrong. Profile first. Optimize second.

Context Beats Complexity

Global Google Trends looked fancy but was useless for Malaysian editors. Suggested topics like "Taylor Swift tour" and "US elections."

Once I added regional intelligence (Malaysia, Singapore, Indonesia, Thailand), engagement doubled overnight. Editors started using it daily.

Lesson: Build for your specific context, not generic use cases. Localization isn't optional — it's the whole point.

What I'd Build Differently Today

If I started this project again tomorrow, I'd:

1. Build the "Oh-Shit" buttons first — disaster recovery before features. The Purge tool and Clear Caches should've existed on day one.

2. Start with WebSub — no polling phase at all. Why waste three weeks when the answer is in the docs?

3. Add more regional intelligence APIs — maybe TikTok trends, Reddit signals, Instagram hashtags. Content intelligence is too valuable to limit to one source.

4. Log everything from day one — debugging is 10x easier when you have forensics. I added comprehensive logging after the second crisis. Should've been there from the start.

5. Write the "How to Steal This" guide while building — documentation debt is real. Writing it all down months later meant I'd forgotten the small decisions that mattered.

But honestly? The current system works so well that "v2" keeps getting delayed. When something just works, you leave it alone.

How to Steal This (Ethically)

You don't need to rebuild YouTube.
You just need to respect what it already gives you.

Start simple:

1. Treat YouTube as your database — one-way sync from YouTube → your system
2. Use WebSub for updates — push beats poll every time
3. Cache smartly — multiple layers, surgical invalidation
4. Build admin tools early — you'll need them sooner than you think
5. Layer automation only where needed — don't over-engineer

The best engineering often looks boring — until it saves you RM500K a year.

If You're Reading This as a Hiring Manager…

You don't need another SaaS subscription.
You need people who can engineer their way out of expensive problems.

That's what this CMS represents — not a product, but a mindset:
designing for performance, sustainability, and reality.

This system proves that:

• Small teams can build enterprise-grade infrastructure
• Smart architecture beats expensive vendors
• Open platforms (YouTube) can outperform proprietary solutions
• Frugality and innovation aren't mutually exclusive

Live system: freemalaysiatoday.com/videos
LinkedIn: linkedin.com/in/mohit5783
Email: mohit5783@gmail.com

Written by Mohit Shrivastava — Head of IT at Free Malaysia Today. Building systems that serve millions, one API call at a time.