TadFuji/ai-news-bot
Serverless AI news curation pipeline: 36 RSS feeds → Gemini 3 Flash dual-pass → Japanese morning brief. Delivered via GitHub Pages, LINE, and X.
What's novel
Serverless AI news curation pipeline: 36 RSS feeds → Gemini 3 Flash dual-pass → Japanese morning brief. Delivered via GitHub Pages, LINE, and X.
Code Analysis
12 files read · 4 roundsA fully automated serverless pipeline that collects AI news from 66+ RSS feeds, uses Gemini AI to translate/score/curate articles into a Japanese morning brief, and distributes via GitHub Pages, LINE, and X (Twitter).
Strengths
Exceptional production-grade error handling with layered guardrails (URL hallucination prevention, source rebalancing, 10-article guarantee, thread-safe extraction, socket-level timeouts) and a genuinely thoughtful dedup algorithm using Japanese character bigrams with complete-linkage clustering. Every non-obvious engineering decision is documented with rationale comments explaining why, not just what.
Weaknesses
The core 'intelligence' is delegated entirely to Gemini prompts rather than custom ML logic, and the keyword scoring in Stage 1 is a simple regex match count — the real value is in prompt engineering and guardrail design rather than algorithmic novelty. Some modules (build_pages.py, generate_weekly_column.py) were not deeply verified, and the project has no CI test coverage metrics or type annotations beyond hints.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
174
Contributors
2
Files
149
Active weeks
9
Repository
Language
Python
Stars
1
Forks
0
License
MIT