
Every morning, millions of people skip the news — not because they don't care, but because reading takes time they don't have. NewScriber solves that with a fully automated pipeline that goes from raw headlines to a listenable podcast episode before your first coffee. The system runs on a scheduled pipeline: at 6:30 AM, it scrapes top tech and business headlines using Firecrawl v2, then passes them through Azure OpenAI (Kimi-K2.5) to editorially select the most relevant stories and generate a natural, two-host conversational script. At 7:00 AM, the script is sent to Gemini TTS via OpenRouter, rendered as high-fidelity speech, transcoded with ffmpeg, and uploaded to Cloudflare R2. The RSS feed updates automatically — and the episode is live on Spotify and Apple Podcasts. This happens in parallel across four languages. English, Italian, French, and Bengali each have their own pair of named hosts (Alex & Sam, Sofia & Marco, Marie & Pierre, Nusrat & Fahim), their own editorial voice, their own timezone schedule, and their own podcast channel. Adding a new language requires changing one prompt and one cron block. The backend is a Go webhook microservice orchestrated by n8n, persisted in PostgreSQL, proxied via Caddy, and deployed on GCP. Everything is containerized with Docker Compose. The architecture is modular by design — the WhatsApp delivery layer is decoupled from the news pipeline so either can scale or be replaced independently.
19 May 2026