News Briefer — AI Daily News Briefings for Organizations
A SaaS that delivers AI-curated daily news briefings filtered per organization. A two-pass LLM pipeline selects and summarizes the local stories that matter to each subscriber. Already in use by the UN World Food Programme in Timor-Leste.
Key features
- Per-organization briefing profiles — describe the organization, add scoring keywords, manage recipients; every briefing is personalized against that profile
- Two-pass LLM pipeline: the first pass selects relevant stories for the organization, the second writes a narrative briefing with an editor's note
- Three-stage deduplication memory (article ID, normalized URL, headline similarity) so no organization is told the same story twice
- Daily scheduled runs through n8n with per-subscriber failure isolation — one failed briefing never blocks the rest of the country run
- Full-text searchable briefing archive on Postgres tsvector with sandboxed HTML previews
- Strict plane separation: n8n orchestrates but never touches the database — all data access goes through authenticated API routes with timing-safe secrets
- Rate limiting that fails closed in production, Row-Level Security on every table, hardened security headers
- Admin console with user and role management, including self-demote guards
- ~124 automated Vitest tests across the API surface and pipeline logic
Technologies
About the project
news.briefer.media grew out of the morning-briefing automations I run for consulting clients: every organization wants to know what happened overnight that affects its work, and nobody has time to read ten news sites to find out. The product turns that into a self-serve SaaS. An organization describes what it does, adds the keywords that matter, picks recipients — and every morning a short narrative briefing lands in their inbox, built only from the local stories that actually affect them.
Each briefing comes from a two-pass LLM pipeline: the first pass scores every gathered story against the organization's profile, the second writes the briefing itself — narrative prose with an editor's note, not a list of links. A three-stage deduplication memory (article ID, normalized URL, headline similarity) guarantees the platform never tells the same organization the same story twice, and every briefing lands in a full-text searchable archive.
Architecturally, the workflow plane and the data plane are strictly separated: n8n handles scraping, summarization, and delivery but holds no database credentials — every read and write goes through authenticated API routes with timing-safe secret checks, rate limiting that fails closed in production, and Row-Level Security on every table, all covered by ~124 automated tests.
The platform launched first with a Timor-Leste edition and is already in use by the United Nations World Food Programme (WFP) in Timor-Leste, with more countries and languages rolling out.