AffairsAI
A bilingual current-affairs product that turns resilient public-feed ingestion into daily briefs, validated quizzes, research digests, and offline study tools.
The operating problem
Public news and government feeds are slow, malformed, and independently unreliable, while model providers may fail or be absent.
My contribution
Built collection across 119 registered source variants, cleaning, deduplication, scheduled daily briefs, validated quizzes, bilingual generation, SEO, and an offline-first Expo client.
System shape
The key decision
Per-source aborts and settled batches allow healthy sources to update. Extractive fallback is explicitly labelled instead of being presented as generated analysis.
Reliability by design
The rolling feed caps storage, normalizes Unicode, sanitizes content, and uses provider-compatible model calls with retries and timeouts.
What the source proves
Honest limits
The proposed multi-agent newsroom is planned only. Public cost-bearing routes and citation depth remain open hardening work.