Case study · 2026
InsightLoop
RSS in → structured AI analysis → Postgres → a personal daily Telegram digest.

- Analyze once with Zod-validated LLM output — never re-score the same article.
- Multi-reader personalization on shared analyzed content.
- Production path: Vercel Cron + Supabase on a free-tier stack.
Problem
Most “AI newsletter” demos stop at a chatbot UI. The hard part is a reliable ingest → analyze → deliver pipeline that stays cheap, dedupes sources, and personalizes digests without re-running models for every reader.
Approach
InsightLoop collects RSS feeds, canonicalizes URLs, runs a structured LLM pass (Groq primary with optional failover), stores results in Postgres via Drizzle, and sends card-style Telegram digests. A Next.js dashboard covers Discover, preferences, pipeline health, and admin workflow controls.
Outcome
A portfolio-grade product you can click through live: public Discover, topic onboarding, Telegram connect, and a daily production cron — focused on orchestration, not just prompts.