Available for selected work
← Back to projects

Work / 01

Ongoing system

AI Briefing Workflows

Select AI updates and turn them into concise, verifiable Chinese briefings.

Type
Personal workflow
State
Ongoing personal workflow
Focus
Source-backed Chinese briefings
Workflow diagramTest → Document

01

Output preview

There is no public briefing artifact in this repository yet, so this page shows the structure the workflow uses rather than inventing a result.

Briefing anatomy / structure example

Evidence before confidence.

Structure example only — not real news, a live feed, or a record of a completed run.

  1. 01

    Headline

    State the verified event plainly, without turning an inference into the event itself.

  2. 02

    Summary

    Explain why it matters while staying inside what the source supports.

  3. 03

    Event date

    Record when the event happened, separately from when it was found.

  4. 04

    Source

    Link to an official or original source that a reader can check.

  5. 05

    Verification state

    Show verified, unresolved, or paused explicitly; missing evidence stops the item.

Reusable structure shared by the home page and case study

02

The problem

AI information moves quickly, but more information does not automatically make a briefing more useful. A summary without original sources, actual event dates, or clear boundaries can create confidence that the evidence does not support.

What I built

I am developing a recurring workflow for concise Chinese AI briefings. It starts from original sources, keeps only a small set of important, verifiable items, and records event dates, retrieval dates, sources, and verification states. When evidence is insufficient, it stops; the result stays clear, readable, and traceable.

  • Select a small number of genuinely important items.
  • Build the working format around a title, short summary, actual event date, and verifiable source, while keeping any remaining gaps visible.
  • Prefer official or first-party sources.
  • When reliable same-day material is insufficient, widen the time window explicitly instead of presenting older news as new.
  • Verify local report generation and external delivery as two separate outcomes.

03

How it works

Collect → Verify → Select → Write → Deliver → Record

  1. 01

    Collect

    Gather a focused set of candidate updates from current sources.

  2. 02

    Verify

    Check the original source and the date the event actually happened.

  3. 03

    Select

    Keep only the small number of items that are genuinely worth reading.

  4. 04

    Write

    Create concise Chinese summaries while preserving source boundaries.

  5. 05

    Deliver

    Attempt delivery without treating “generated” as proof of “received.”

  6. 06

    Record

    Preserve the run status for follow-up checks and future deduplication.

04

Reliability before speed

  • Prefer first-party sources.
  • Separate event date from retrieval date.
  • Make uncertainty visible.
  • Verify generation and delivery independently.
  • Stop instead of inventing missing information.

What I learned

Reliable automation is not always running; it knows when to stop.

The hard part is guarding the boundaries: stop when evidence is thin, mark unclear dates, keep checking unconfirmed delivery, and do not repeat completed work. This personal system is still evolving, leaving decisions that can be reviewed and explained after every run.

05

Ongoing

Current status

I continue to improve source selection, exception handling, and delivery verification. The system is deliberately presented as ongoing: the goal is clearer judgment and more dependable boundaries, not a claim of completion.