Show HN: RSS/Atom Feed Summarizer (github.com)

A developer released a spec-driven, asyncio-based RSS/Atom (plus optional Mastodon) feed summarizer that fetches feeds, stores raw items in SQLite, generates AI summaries via Azure OpenAI, groups items into topical “bulletins” (three daily by default), and publishes both responsive HTML and RSS outputs — optionally uploading to Azure Blob Static Website hosting. The project is delivered as a background service with simple quickstart (five commands), Docker Swarm/kata deployability, and CLI modes for one-shot runs, scheduled operation, fetch-only, summarize-only, and upload-only. Core features include conditional fetching (ETag/Last‑Modified), batching, respectful backoff/error tracking, retry + bisection for content-filtered responses, MD5-based dedup when uploading, and timezone-aware smart scheduling. This is significant for AI/ML practitioners because it demonstrates a practical, production-minded integration of LLM summarization into a real ingestion pipeline: prompt templates (prompt.yaml), per-group introductions, configurable retention windows and summary-size caps (time_window_hours, MAX_ITEMS_PER_FEED, SUMMARY_WINDOW_ITEMS), OpenTelemetry observability hooks (optional Application Insights export), and hot-reloadable configuration. Operational controls and safeguards (rate limits, content-filter handling, telemetry opt-out, secrets handling) make it suitable for long-running use and experimentation with different models (OPENAI_API_KEY, DEPLOYMENT_NAME default gpt-4o-mini). The code is MIT-licensed, test coverage and hardening are planned, and the repo serves as a useful reference for teams building reliable LLM-powered feed summarizers.