# Silent for Weeks, Then Every Other Day: How Dan Hill's n8n pipelines publish three video brands from a Raspberry Pi

Dan Hill, a compliance systems and data specialist who works in food manufacturing, runs three video brands as a side project on n8n pipelines hosted on a Raspberry Pi. Seed banks feed idea generators, Gemini agents draft scripts and metadata, a self-hosted voice clone narrates, and render services produce long-form videos, Shorts and TikTok clips. His only human step on two of the brands is approving each script from an email form. One brand, The Food Compliance Desk (TFCD), went silent for weeks because an approval timed out, a switch node matched nothing and n8n recorded a success. He rebuilt TFCD in October 2026 with an automated quality gate in place of the email approval, and its Quick Guides have gone out every other day since 6 October 2026.

**Try this step:** <strong>Make every stage write back to one plan.</strong> Each stage writes to a content plan sheet. Seed banks use least-used-motif selection with deterministic tie-breaks, so ideas don't cluster.


| | |
|---|---|
| **Who** | Dan Hill, compliance systems and data specialist in food manufacturing (atechstudio.co.uk). GitHub danatechstudio; YouTube @theninthvolume and @TheFoodComplianceDesk; TikTok @theninthvolume; all linked from his site, verified 2026-11-20 |
| **Problem** | Running a one-person video operation where the human only picks ideas and approves scripts, while pipelines fail quietly |
| **Stack** | n8n (seven workflows per brand), Gemini (agents for scripts and metadata, plus hooks, scene images and thumbnails), Chatterbox self-hosted voice cloning, Python and ffmpeg render services, Google Sheets and Drive, YouTube, Buffer for Shorts to YouTube and TikTok, KDP PDF rendering, Docker on a Raspberry Pi 5 |
| **Confidence** | Medium (first-person build notes; channel counts checked; no audience or revenue results stated) |
| **Last verified** | 2026-11-20 |

## The problem
"Content pipelines break in quiet ways", he writes. "Approvals time out without anyone noticing, renders fail while the run is marked as a success, and the same story gets regenerated again and again." He built the system to find out "how far a one-person content operation can go when the human only does the parts that need taste: picking ideas and approving scripts."

## The blueprint (as he describes it)
1. **Make every stage write back to one plan.** Each stage writes to a content plan sheet. Seed banks use least-used-motif selection with deterministic tie-breaks, so ideas don't cluster.
2. **Approve scripts from an email form.** For The Ninth Volume and The Ninth Archive, a Gmail send-and-wait form offers Approve, Request changes or Decline, with one bounded revision round. Approval is a hard gate.
3. **Count a post only when the platform confirms it.** Long-form goes straight to YouTube. Shorts go to YouTube and TikTok through Buffer's GraphQL API. A row is marked posted only once a post ID comes back, so a failure retries the next day by itself.
4. **Reuse finished work instead of regenerating it.** Voice and render steps reuse output by content key, so a retry carries on from where it stopped. His audit had found one Short regenerated 17 times.
5. **Take facts from reference files, not the model.** The TFCD rebuild moves each item through ten stages, from research to publish. BRCGS clause numbers and Copilot tier facts come from versioned reference files with sources. QC rejects any clause the reference doesn't contain and allows one automatic revision before the item is flagged to him.
6. **Publish the landing page before the video.** Every finished TFCD item gets a page on his site with the worked example, prompts and references. The page exists from the end of production, so the link in the video description works on upload day. A daily n8n check loads every page and emails him if one fails.
7. **Repackage the approved work.** An anthology check queues a KDP print volume once about 200 pages of approved stories exist.
8. **Protect the box.** Uncapped voice inference starved n8n's event loop and crashed it, so containers now have CPU caps and shares, load-tested under a 40-minute inference job. Media files go to Drive and only their references travel through n8n.

## Stated results
- **Publishing:** "Distribution is live on YouTube and TikTok." TFCD published its first rendered video in July 2026. Since 6 October 2026, after the rebuild, its Quick Guides have gone out on YouTube every other day. The latest on his TFCD page is dated 19 Nov 2026.
- **The silent stretch:** TFCD "had gone silent for weeks" because "when an approval timed out, the switch node matched nothing, and n8n recorded a success."
- **Scale on his homepage (counted 21 Oct 2026):** 109 n8n workflows, 66 on a schedule or trigger, and 16 services in Docker on one Raspberry Pi 5 with 8 GB of RAM. These cover all his side projects, not only the video brands.
- **Audience (checked on YouTube, 2026-11-20):** The Food Compliance Desk showed 3 subscribers and 37 videos. The Ninth Volume showed 32 subscribers and 99 videos.
- **Views, subscriber growth or revenue:** None stated.

## Caveats
- Publishing on schedule is the outcome he reports. The channels are small, and he gives no views, watch time or revenue.
- He does not say how many weeks TFCD was silent, or how the fix for the unmatched switch node relates to the October rebuild. The page lists both.
- TFCD's Quick Guides show Microsoft Copilot doing compliance tasks. Copilot is the subject of the videos, not part of the pipeline.
- His site offers ATechStudio services for small businesses, and the TFCD page links a free audit app with a Ko-fi tip link.
- Handles: GitHub danatechstudio; YouTube @theninthvolume and @TheFoodComplianceDesk; TikTok @theninthvolume (all linked from atechstudio.co.uk). No verified personal X, Bluesky, Threads or Instagram profile was found.

## FAQ
### What does Dan Hill still do by hand?
For The Ninth Volume and The Ninth Archive, he approves each script from an email form. For TFCD, an automated quality gate replaced that approval in October 2026, and items are flagged to him after one failed automatic revision.

### Why did his TFCD channel go silent?
An approval timed out, the switch node in n8n matched nothing, and n8n recorded the run as a success. He found the cause and lists it on his project page.

### How does he stop the pipeline from wasting renders?
Voice and render output is reused by content key, so a retry continues instead of regenerating. His audit had found one Short regenerated 17 times.

### How big are the channels?
On 2026-11-20 The Food Compliance Desk showed 3 subscribers and 37 videos, and The Ninth Volume showed 32 subscribers and 99 videos. He states no views or growth figures.

## Sources
- https://atechstudio.co.uk/work/content-pipelines (primary: "AI content pipelines: The Ninth Volume, The Ninth Archive, TFCD", live Jul 2026 to present, checked 2026-11-20)
- https://atechstudio.co.uk/ (homepage with workflow and service counts)
- https://atechstudio.co.uk/tfcd (TFCD library and latest Quick Guide)
- https://www.youtube.com/@TheFoodComplianceDesk and https://www.youtube.com/@theninthvolume (channel counts, checked 2026-11-20)

---
*Want agents like these working your distribution? [Get It Built](/get-it-built). Get new teardowns weekly: [subscribe](/subscribe) (double opt-in).*