Discernion

A news intelligence product that reads 300+ publishers with a multi-model AI pipeline.

Our own product

Discernion home page: a featured story headline with a “why it matters” note beside a list of regional news stories.
Discernion (desktop)
Discernion home page on a phone, with the featured story and its summary.
Discernion (mobile)
Type

AI news intelligence

The problem

AI news tends to be either a raw feed or a chatbot in a banner. Readers who want the point of a story, and a way to check it, get neither.

What was built
  • An ingestion pipeline that reads 385 sources from 314 publishers every two hours. It feeds more than 25 desks, from AI, tech and finance to country desks for Pakistan, India, Germany, Africa and the Middle East.
  • AI agents turn each story into a gist, a note on why it matters, key points, a plain-language explanation, a deeper analysis and an upside and downside view. Each one links to the original reporting.
  • A reading site with search, topic pages, trending, an RSS feed and a newsletter, plus an admin panel where a person can edit or pull any story.
  • Mobile apps for iOS and Android with push notifications and offline reading.
Stack

Next.js and TypeScript on Cloudflare, Supabase (Postgres) for data, React Native for the mobile apps.

What makes it hard

Keeping machine-written summaries honest. Agents may only use facts from the source article, and paywalled, empty or off-topic pages are skipped instead of guessed at.

The same story arrives from many publishers. We catch duplicates by URL, by title similarity and by a fingerprint of the text.

Running a 30,000-article site on free-tier infrastructure. Search, tag pages and crawler traffic once overloaded the database, so we bounded those queries, shrank what crawlers can reach, and stopped caching a failed query as a missing page.

Outcome

More than 39,000 stories published as of 7 October 2026, each linked to its source.