AI Newsletter Automation Pipeline

Publishing a consistent newsletter manually takes 2-4 hours per edition. This pipeline automates the entire process from content sourcing to delivery. **Pipeline stages:** **Stage 1: Scrape** — Fetches articles from a configurable list of RSS feeds. Supports up to 50 sources. Filters by publication date (configurable lookback window, default 24h) and deduplicates by URL and content similarity hash. **Stage 2: Rank and Select** — Scores articles by engagement signals (where available), source authority weight, and keyword relevance to your topic list. Selects the top N stories per section (configurable). **Stage 3: Summarise** — Calls Gemini Flash to produce a 2-3 sentence summary for each story. Prompt engineered for factual, neutral summaries without hallucination padding. Includes a fallback to the article's meta description if the full content is paywalled. **Stage 4: Format** — Assembles the final newsletter from a Handlebars HTML template. Sections are configurable (e.g., Top Stories, Tools & Releases, Research Highlights, Jobs). Inline CSS for email client compatibility. **Stage 5: Send** — Sends via Resend API with batch sending for large subscriber lists. Tracks opens and clicks via Resend analytics. Handles unsubscribe links automatically. **Configuration file** (`newsletter.config.ts`): source URLs, section definitions, send schedule, subscriber list ID, from address, and AI summarisation prompt. **Cron setup:** Vercel cron config included. Also works with GitHub Actions scheduled workflow.