Lumen News Bot: AI-Curated News Briefings on Telegram

The problem

Staying on top of relevant news without drowning in noise is a daily tax for anyone in tech, business, or innovation. Generic newsletters arrive too late and cover too much. RSS feeds require manual triage. The result: either information overload or willful ignorance.

The goal was to build a personal news assistant that does the filtering work automatically — one that learns what matters to a specific user and delivers it, without friction, on a channel they already use every day.

What we built

Lumen is an AI-powered Telegram bot that aggregates content from multiple sources — blogs, newsletters, feeds, and publications — and uses Claude to curate only what's relevant for each user.

The pipeline works like this:

1. Aggregation — n8n workflows pull from 15+ configured sources on a schedule
2. Curation — Claude reads each item and scores relevance against a user profile stored in Notion
3. Delivery — Telegram bot formats and sends personalized briefings, respecting quiet hours and user preferences
4. Feedback loop — user reactions (👍/👎) update the Notion profile, improving future curation

The system runs entirely serverless. No app to install. No dashboard to check. The briefing arrives; you read it or skip it.

Key decisions

Why Telegram instead of email or WhatsApp? Telegram's Bot API is fast, free of per-message costs, supports rich formatting, and doesn't require a business account. For a proof of concept targeting individual users, it removes all the activation friction.

Why n8n instead of a custom scheduler? n8n gives us visual workflow editing, built-in error handling, and connectors to dozens of sources without writing a scraper for each one. For a project that needed to move fast, it was the right tradeoff.

Why Notion as the user profile store? The user profile needed to be editable by a non-technical user and readable by Claude in a single API call. Notion pages are structured enough to parse reliably and simple enough to edit in plain text.

Results

Lumen has been running in production for the founder's own use since January 2026. It surfaces 3–5 relevant items per briefing from a pool of 50–80 daily articles — a signal-to-noise improvement that would take 20–30 minutes of manual triage to replicate.

More importantly, the architecture proved out a pattern we now apply to client chatbot projects: an orchestrator that reads a stored profile, evaluates incoming data against it, and delivers structured output over a messaging channel with human-readable formatting.