[officehours 46] Automate Doc Understanding and Generation w/ Claude Opus + Google Gemini + n8n Workflow

​In officehours 45, Colton showed you how to track different non profits and whether they were mentioned in the news and updated Hubspot with a task to check it out. This week we will focus on understanding and generating documents predictably. That's right PREDICTABLY!

​The honest version of "just use AI to write your documents."

​If you've actually tried to generate real documents with ChatGPT or Claude, you know the pitch and the reality don't match. You start a Project, upload your files, get a decent draft — then you need another version, so you save a new file. Then another. Now you're managing a folder of near-duplicates by hand, re-uploading context every session, and wondering why the "memory" doesn't remember the thing you needed. Cowork makes the file-juggling a little saner — until you watch your credits evaporate halfway through a real workload.

​So how do you do this for real — repeatably, at volume, without babysitting a chat window or burning through credits?

​This session, we build the open-source way.

​We'll show you how to stand up a document generation and understanding workflow using tools that are free to run and free to keep:

• ​n8n as the automation backbone — the workflow that orchestrates everything

• ​SKILL.md files to encode how a document should be built, so quality is consistent every run

• ​OpenRouter to route across ChatGPT, Claude, and Gemini from one place — pick the right model for the job, control cost

• ​MCP servers — a handful of readily-available open-source tools that give your workflow real document capabilities (read, parse, generate, file)

• ​A little DevOps / no-code glue to wire it together and connect the chat models you already use into the pipeline

​By the end you'll understand the difference between using an AI to make a document and building a system that generates documents on demand — and why the second one is the only thing that scales.