agent workflow
AI news digest workflow
An admin-reviewed workflow for collecting trusted tech sources and drafting Vietnamese digest and vlog-script content.
Pipeline
7 steps
Publish gate
Admin
Sources
Trusted
problem
Automated content is risky when it crawls broadly or publishes directly. The workflow needed source provenance, dedupe/ranking, draft review, and a hard approval gate.
outcome
The news agent fetches configured sources, extracts and ranks articles, generates digest and vlog drafts, then keeps publishing behind explicit admin review.
What it proves
Trusted source pipeline instead of arbitrary web crawling.
Article provenance, hashes, run logs, and step logs are persisted.
Drafts remain pending until an admin approves, edits, or rejects.
Public digest pages render only approved published items.
build notes
Workflow shape
The agent moves through fetch, extract, dedupe, rank, digest draft, save draft, and vlog-script generation. Each phase is designed to leave enough trace data for review and debugging.
Safety model
The project avoids arbitrary crawling, keeps generated content in draft state, and requires admin approval before anything becomes public.
Product fit
The workflow supports the personal brand/blog side of the portfolio: it creates raw material for regular technical commentary without pretending generated drafts are finished writing.
next case study