Our own product2026Türkiye Anlık
An AI news service running on an automated publishing pipeline.
- Project
- Our own product
- Our role
- Product · Full stack · Content pipeline · Mobile
- Duration
- 4 months
01 / Behind the product
A real need. A considered solution.
Challenge
On a news site speed is everything, but hand-written content neither scales nor stays consistent. Handing it to a model brings two risks: unverified text going live, and the same story being rewritten every cycle. The monthly cost target was 80–120 dollars, so every model call had to be visible.
Approach
The pipeline runs on a Cloudflare Worker (Hono) in five stages: a radar scans sources every 15 minutes; a selection step hands the model the last 72 hours of story clusters so it picks only new events or genuine developments, and a signal that attaches to an existing cluster becomes a revision rather than an article; a writer produces 400–600 words with at least two independent sources; a separate verification pass checks factual consistency, headline exaggeration, language and source integrity before publishing (PASS / FIX / FAIL); and at the start of every cycle the previous cycle’s articles are compared against fresh search results, with corrections published as revisions with history.
Every run is logged with stage, duration, tokens and estimated cost. The site is Next.js 16, the mobile reader Expo; schema validation lives in a shared package. Push respects an importance threshold, categories and quiet hours.
Result
turkiyeanlik.com and api.turkiyeanlik.com are live; web deployment is automated with GitHub Actions, while deploying the API that prints live news is deliberately kept under human supervision. One command stops or starts the whole pipeline; when stopped no model query is made and the site stays up. A weekly GEO audit runs as its own cron.
02 / Inside the product
The experience, on screen.
Open a screen to explore the details.03 images
03 / Scope & details
The details make the experience.
- 01
Scheduled radar plus a batch publishing pipeline
- 02
Site and mobile reader from one API
- 03
Schema validation in shared packages
- 04
Automated deployment
Technology
Next.js / Cloudflare Workers / Supabase / Expo
INSIDE THE PRODUCT
The engineering behind the interface.
Systems we built and concrete problems we addressed in this project.
Duplicate-prevention memory
story_clusters keeps the last 72 hours of cluster keys; the selection step hands the list to the model and asks only for new events or genuine developments. A signal that attaches to an existing cluster becomes a "development" revision of that article rather than a new one.
Independent verification before publishing
A model call separate from the writer, holding the scan stage’s raw signals and source URLs, checks factual consistency, headline-body agreement, language and source integrity. PASS publishes, FIX publishes the corrected text, FAIL rejects the article and tries the fourth signal once.
Transparent correction history and cost
article_revisions stores a full copy of the previous content and the reason ("initial", "development", "correction"); pipeline_runs records each stage’s duration, tokens and estimated cost. The device table carries the importance threshold, categories and quiet hours.
Your idea could be next.
What could we build for you?
From a first release to a growing platform. Let’s plan the design, development and launch together.
From Türkiye to the US, UK, Australia and everywhere in between.