A sheet row in, a brainrot reel out

The problem

Tech explainers are dry, and editing them takes hours

Short videos that explain AI and tech terms are useful, but most are a talking head reading definitions. The brainrot format, two characters chatting over gameplay footage, is built to keep people watching.

Making one by hand means recording two voices, cutting every line to the footage and adding the speakers, every single time.

Why I built it

Could a channel run on a spreadsheet?

I wanted to see how far automation could go: if I only chose the topics, could a channel make its explainers on its own? Peter in Tech was the test, covering AI agents, RAG, MCP, vector databases and tech news.

How it works

From a sheet row to an MP4 in five steps

n8n, a workflow tool running on my laptop, is the conductor. It reads the sheet, calls each service in turn, waits for the render and writes the result back.

1. You

   Pick a topic and paste Claude's script into the sheet.

2. n8n

   Runs all 16 steps, one reel per run.

3. Script Claude

   A two-voice dialogue of 18 to 30 lines: S asks, P explains. Plus a caption for the post.

   SWait so everyone keeps saying AI agents like it means something different every time. What actually IS an AI agent?

   POkay so imagine you hire an intern. Not a very smart intern. But they can use a computer, browse the web, send emails, and make decisions.

4. Queue Google Sheets

   One row per reel: topic, script, caption and status. n8n takes the next row marked ready, and marks it completed with the video link when it's done.

   Reel	Topic	Words	Status
   002	What are RAGs	224	completed
   003	What are MCPs	240	completed
   005	Fine-tuning vs RAG	241	completed
   008	CORS error explained	299	completed
   011	Kunal Shah leaves CRED	403	completed

5. Voices FFmpeg · Coqui

   Coqui, an open-source speech model on my laptop, clones two voices from short reference clips and speaks every line as S or P: one audio file per line.

   

6. Visuals Cloudinary

   One of 11 Minecraft parkour clips, picked at random for each reel, plus the artwork for S and P.

7. Render Remotion · FFmpeg

   Remotion builds the video in React, scene by scene. Each scene lasts exactly as long as its line of audio and shows whoever is speaking. FFmpeg encodes the MP4, and the sheet row is marked completed.

   

What happened

11 reels, then a pause

The workflow rendered 11 reels in June 2026, seven of them in under an hour. It worked, but it wasn't fully hands-off: topics and scripts still went in by hand, and every service ran on my laptop, so the laptop had to be on for a reel to get made. The channel is paused for now.