tayf

Aynı haber, farklı dünyalar.

The same story, different worlds. A Turkish news bias analyser that ingests 144 outlets across the political spectrum, clusters the ones covering the same event, and shows you how each side framed it.

Current state Side project · in dev
Built with A friend, on the side. Two of us.
Not an agent tool A different problem entirely — it is here because we are building it, not because it fits

144 outlets. One event. Seven incompatible accounts of it.

Turkish media splits cleanly along known lines, and everyone already knows which paper they trust. The interesting question is not who is biased — it is which stories a whole side of the spectrum simply did not run.

A story, by who ran it

Every card leads with this bar rather than a grid of mastheads. The width is how much of the coverage came from each bucket. The hatched slot is the finding.

Coverage distributionexample
Pro-government31% State media18% Nationalist14% Islamist-cons.12% International15% Opposition10% Kurdish outlets0%

That empty slot is a blindspot. A story carried by six buckets and ignored by the seventh, surfaced on the card instead of buried in a filter nobody opens.

How it works

RSS in, clusters out, continuously. Nothing here is a language model doing the clustering on vibes.

Ingest

144 feeds, always on

RSS from outlets spanning the full spectrum, pulled on a schedule rather than on demand, so the archive is there before anyone asks a question of it.

Cluster

Three methods, not one

An ensemble decides which articles are covering the same story. One method alone either splits a story into five or welds two unrelated ones together.

Surface

Blindspots and surprises

Two flags: stories only one side ran, and the more interesting case — an outlet breaking ranks with the bucket it normally sits in.

Built with
Next.js 16 App RouterReact 19TypeScript 5 Tailwind 4 · OkLCh tokensshadcn/uiSupabase Postgres pgmq queuesEdge Functions · Deno 2pg_cronVercel

What it isn’t

Worth saying plainly, given what it touches.

Not a verdict

It doesn’t tell you who’s right

Bias buckets describe where an outlet sits, not whether a given article is true. The tool shows you the spread and the gaps; the reading is still yours.

Not a Mirako tool

A side project, with someone else

It has nothing to do with coding agents and it is built with a friend rather than by the studio. It is listed here because it is real and we are building it, and leaving it off would be the tidier kind of dishonest.

Where it’s at

In development, on the side, at the pace two people manage around everything else. The ingest and clustering run; the reading experience is the part still moving.

Ground News exists and is good. This one is for a media landscape that one does not cover, in a language it does not read.

Read Turkish news?

If you follow this stuff and think a bucket is drawn wrong, that is genuinely useful and hard to get.

hi@mirako.computer