Proof capsule
Media Monitor
Operational AI/data automation system
An AI/data intelligence pipeline for turning raw news inputs into structured briefs, drafts, and human-reviewable outputs.
Problem
News monitoring becomes fragile when sources, scraped text, editorial notes, generated drafts, and publishing surfaces are mixed into ad hoc folders or one-off scripts. The system needs stable boundaries: acquisition, enrichment, editorial generation, contracts, handoff, and public snapshots.
What I built
A modular monorepo pipeline that ingests news references, groups them into digest artifacts, enriches article text, generates editorial briefs/drafts, validates contract buses, and exposes compact handoff/public surfaces.
What this proves
- Design inspectable AI/data workflows across acquisition, enrichment, editorial generation, and publishing boundaries.
- Separate LLM runtime output from stable contract buses and human handoff surfaces.
- Expose fallback visibility and operational status instead of hiding fragile assumptions in scripts.
Module boundaries
news_acquire
Owns raw acquisition boundaries such as RSS dumps, digest maps, digest JSONLs, master references/indexes, acquisition quarantine, and buses including news_ref.v1 and news_digest_group.v1.
news_enrich
Fetches and normalizes article text from known references, emits schema-valid scraped_article.v1 records, and keeps enrichment status observable for manual, batch, worker, and recovery paths.
news_editorial
Turns digest runtime inputs into editorial briefs, drafts, and handoff indexes. The stable contract is not raw LLM output; it is schema-valid buses and a compact decision surface for human review.