Architecture

Architecture

Daily Bits By AI v2 is designed as an autonomous AI signal engine.

The workflow starts with monitored sources, converts raw source items into candidate signals, scores those candidates, selects useful topics, generates posts, verifies claims, publishes to WordPress, and logs the workflow run.

Pipeline

1. Source Collector

2. Candidate Generator

3. Deduper / Clusterer

4. Scoring Agent

5. Editorial Selector

6. Research Agent

7. Writer Agent

8. Verifier Agent

9. Publisher Agent

10. Feedback Agent

Tools

  • WordPress for the public website and publishing destination
  • Airtable for operational data
  • n8n for workflow orchestration
  • Multiple LLMs for generation, scoring, writing, and verification
  • RSS, APIs, GitHub, research feeds, and official changelogs for source ingestion

Data Flow


Sources
  -> n8n Source Collector
  -> Airtable Raw Source Items
  -> LLM Candidate Generator
  -> Airtable Candidate Signals
  -> LLM Scoring Agent
  -> Selected Candidate
  -> Research and Writer Agents
  -> Verifier Agent
  -> WordPress Publisher
  -> Airtable Published Posts and Run Logs

Verification Layer

The verifier checks drafts for unsupported claims, incorrect dates, model or company name mistakes, pricing or availability claims without sources, overstated conclusions, missing caveats, and weak source links.

Drafts can be approved, revised, or rejected. Rejected claims are stored so the system can learn from them.

Why This Exists

The project is designed to explore agentic AI workflows in practice. The content is useful, but the system behind the content is the main artifact.