One Mac, running multiple content channels on autopilot

The problem

Running content at scale usually takes a whole editorial team: find topics, translate, rewrite, format, publish, watch the numbers. Point a raw AI at it and you get obviously-machine text — sometimes with the facts wrong. Distill's goal: automate the entire line while holding the bar at "reads like a human, and doesn't get things wrong."

How it works

1. 01
   Pick

   Selects topics from dozens of frontline RSS / radar sources — deterministic code de-dupes and filters off-topic items, no LLM guessing.
2. 02
   Fetch source

   Pulls the full article as the single source of truth; skips the cycle if it can't — never fabricates from a headline.
3. 03
   Rewrite

   Rewrites in Traditional Chinese in its own words, applying a 14-point "de-AI" checklist and proven first-line hooks — not machine translation.
4. 04
   Fact-check

   Every place name / proper noun must appear in the source, or the draft is rejected and rewritten — stops AI hallucination.
5. 05
   Publish

   Auto-schedules to each channel; images, source comments, length and formatting all handled automatically.
6. 06
   Sediment

   Each post also becomes a page on an owned website — a long-term, Google-indexed asset (the site you're on).
7. 07
   Feed back

   Collects reach data daily; weekly it distills "which opening line works" back into strategy, so the system keeps getting sharper.

Why it's not "just ask ChatGPT"

The difference is engineering. This system treats "don't get it wrong" as a reliability problem, not a roll of the dice:

• Fact-check gate
  Place names / proper nouns must match the source or the draft bounces. Once caught a piece that placed an Osaka shop in Tokyo.
• Four-layer publish defense
  Even if the writer crashes mid-run, no empty image-only post ships — four independent guardrails.
• De-AI self-audit
  A 14-point checklist strips canned phrasing and filler so readers can't tell a machine wrote it.
• Data-driven
  First-line structure isn't a guess — it's distilled from real reach data (a 20× spread between best and worst).

Who built it

I'm Hsiao-Yu Tung, a Principal Data Engineer. My day job is exactly this: taking social-platform data end-to-end into a warehouse and keeping it running on schedules and models. Distill applies the same engineering to content production — pick, rewrite, verify, publish, sediment, feed back — where every stage is maintainable automation, not a one-off demo.