Local LLM Blog Digest

2026, Apr 30

I built a small system to generate a daily blog digest using a local LLM and Notion. Here’s the technical walkthrough—kept simple and practical.

1. RSS ingestion

Defined a feeds.yaml with sources and limits.

Used feedparser to pull entries and filtered:

  • only recent posts
  • limited per day

2. Content extraction

For each post:

  • fetched the article URL
  • extracted clean text using newspaper3k

Key constraint:

  • trimmed content (~1500 chars) to keep LLM input bounded

3. Local LLM summarization

Instead of external APIs, used Ollama:

  • model: gemma4:e4b
  • invoked via HTTP (/api/generate)
  • prompt enforced structured JSON output

Handled common issues:

  • markdown-wrapped JSON
  • inconsistent formatting → cleaned before parsing

🔧 Prerequisite: Running the model

Before summarization works, you need a local model running via Ollama.

ollama run gemma4:e4b

This starts a local server at:

http://localhost:11434

The script sends prompts to /api/generate and receives summaries.

4. Notion integration

Used Notion API via notion-client to:

  • create one page per day
  • structure content with headings + paragraphs
  • include clickable links

Important fix:

  • switched to timezone-aware datetime (datetime.now().astimezone()) to avoid incorrect dates

5. Configuration + secrets

  • .env for API keys (python-dotenv)
  • CLI args for feeds + database ID
  • avoided hardcoding

6. Automation

Wrapped execution in a shell script:

  • uses absolute paths
  • logs per run (timestamped)
  • avoids reliance on shell environment

Scheduled with cron:

  • learned macOS quirks (permissions, environment isolation)
  • fixed via explicit paths + permissions

Added system wake using pmset to ensure execution window.

7. Reliability fixes

  • handled LLM output variability
  • added logging for observability

8. Packaging

  • created requirements.txt
  • documented setup in README
  • pushed to GitHub

Final system

RSS → extract → local LLM → structured output → Notion

Runs locally, no external LLM dependency, and produces a clean daily digest with minimal maintenance.

Key lessons:

  • control input size early
  • don’t trust LLM output format blindly
  • cron on macOS needs explicit handling
  • keep config and secrets separate

You can find the core code repository here: https://github.com/joshi-aparna/local-llm-blog-digest



# tech