Case Study: How We Built a Fully Automated News Portal

Vyraa.com is a digital entertainment news portal we built and run inside Nyza Creations. It publishes 24 articles per day across Bollywood, Hollywood, and music — all AI-generated, fact-checked, and distributed across Instagram and Facebook. No human writes any content. We invest about 30 minutes per month on it.

Here's exactly how it works.

The architecture in one sentence

A cron job runs hourly, decides what to write about, generates an article + a hero image, writes social captions, publishes everywhere, and logs everything to a Postgres database.

Step 1 — Topic discovery

The first decision is "what do we write about right now?" We use a combination of:

• Serper API for what's trending on Google News in entertainment categories
• YouTube Data API for new music videos with growing view counts
• Internal balancing logic that ensures we cover Bollywood + Hollywood + music every day, not just whichever has the loudest news

The trending query gets ranked by recency, search volume, and our existing coverage. The bot picks one topic per run.

Step 2 — Research

For the chosen topic, we run a structured research pass:

• Top 5 articles from major outlets on the topic
• The Wikipedia entry for any named people or films
• Recent social media discussion volume
• Any YouTube video that's directly relevant

All of this gets stuffed into a context window with explicit instructions on what to use vs. ignore. The bot is told to write a 600-word original article that synthesizes (not copies) the sources, with a hook and a satisfying ending.

Step 3 — Drafting

We use OpenRouter to route to Gemini 3.1 Flash for the main article body (great voice, cheap) and Claude Haiku 4.5 for the headline (better hooks). Total cost per article: roughly $0.02.

The bot generates:

• Article body (~600 words)
• SEO meta title and description
• 3-5 internal-link suggestions
• A hero image prompt for the image generator

Step 4 — Hero image generation

Each article gets an original 1200×630 hero image generated by Imagen 4 Fast at $0.02/image. The image is also resized to 1080×1080 for Instagram and a 16:9 thumbnail for the article.

We picked Imagen over Gemini 3.1 Flash Image because Imagen is dramatically cheaper at our volume even though Gemini 3.1 has slightly better quality. At 24 articles/day, those pennies matter.

Step 5 — Publishing

The article + image gets inserted into a MySQL database that powers vyraa.com. Next.js statically generates the article page on next build. Clean URLs (`/article/title-slug`) are rewritten via `.htaccess` so social platforms get clean URLs to share.

Critical detail: OG meta tags are server-side rendered (via PHP) so Facebook and Twitter previews work. We learned this the hard way after early articles got shared without preview images.

Step 6 — Social distribution

The same cron run that publishes an article also posts it to:

• Instagram (@vyraanews) — square hero image + caption + hashtags
• Facebook (Vyraa News page) — link preview using the OG tags

Both use the Meta Graph API. The bot writes platform-specific captions (Instagram is more conversational, Facebook is more headline-driven) using the same article context but different prompts.

We're not on TikTok yet because video generation at the quality + cost we'd need isn't quite there yet. Coming in 2027 with whatever the next round of video models brings.

Step 7 — Monitoring

A Slack webhook fires after every successful publish with the headline and the live URL. A second webhook fires on any error (failed image gen, social posting failure, DB write failure). Most days I get 24 success messages and zero errors.

We log every article generation to a Postgres table with the topic source, sources used, model versions, token spend, and final URL. This is what lets us audit quality and tune the pipeline.

What it cost to build

Total build time: about 60 hours of engineering across 3 weeks. Total ongoing cost: ~$40/mo of API spend + $12/mo for the droplet hosting it. Revenue from ads: about $200/mo and growing. The system pays for itself.

What it would have replaced

If we ran Vyraa with a human team, you'd need:

• One writer (~$4,000/mo) to write 24 articles/day at survival quality
• One social media manager (~$3,500/mo) to handle posting and engagement
• One editor (~$2,000/mo part-time) for quality control

That's $9,500/mo of labor. The AI pipeline does 80% of what those humans would do for $40/mo. The remaining 20% — strategy, audience building, monetization — is what humans should be doing anyway.

What's *not* automated

We made conscious choices about what to keep human:

• Major editorial decisions — what category to add next, when to take down a story, response to a legal letter
• Monetization strategy — ad placements, sponsor relationships
• Brand voice changes — when to pivot, when to add new platforms
• Crisis response — anything that would require a human-judgment decision

All of these are infrequent and high-stakes. AI shouldn't make them.

Could you copy this for your business?

Yes — if your business is content-heavy. The exact pipeline transfers to:

• Real estate listings + market-update articles
• E-commerce product launches + buyer guides
• Legal/medical practices doing educational content
• Any local business with a "news and events" angle

The trickiest part is not the technology — it's the quality control loop. You need to read at least a few outputs every week and tune the prompts when they drift.

If you want to build something like this for your business, take the AI Readiness Audit and we'll tell you whether a content pipeline is your best play or if there's a bigger win elsewhere.