Show HN: Clarion – AI system that rejects 97% of news into a high-signal digest

Hey HN — I built this because opening the news started to feel like drinking from a firehose. The signal-to-noise ratio keeps dropping, and most feeds optimize for outrage, not insight.

Clarion is a small AI editor that reads thousands of articles weekly, filters hard (~3% make it through), and surfaces only progress-focused stories. The goal: a clear weekly snapshot of what’s actually moving forward, without the doomscroll tax.

It runs on a simple stack: Vite + TS frontend, Supabase backend, automated ingestion, and a strict scoring pipeline. Built for myself at first, but it’s been resonating with founders, execs, and other busy people.

Stack • Ingestion: Lambda workers + Supabase Edge • Scoring: Custom AI pipeline (Gemini 2.5 Flash-Lite, Gemini 2.5 Pro, Claude) • Frontend: Vite, React, TypeScript • Infra: Supabase Postgres on AWS

Would love feedback on scoring, UX, or anything that feels off.