What I learned building my first AI automation (using an open source LLM)
How to get started
Hi friends,
I built my first end-to-end AI automation using a local LLM.
AI automations using n8n are trending for a reason.
It allows non-techies (like me) to stitch together tools and AI and take automations to whole other level.
I have to admit that I fell into the trap of doomscrolling tutorials and did not have the time (rather didn’t make the time) to build until one random Saturday.
It was a intense learning curve, but you’ve got to start somewhere.
You will get the full explanation of why and what I built here.
This post is for anyone trying to build something useful and real with AI for the first time, but feel overwhelmed to start.
The use case
I came up with a very boring and simple use case.
Extract and populate details about from a business lead into a Google sheet to track progress.
The goal was simple:
Every time someone emails, I want to extract their Name, Company, Email, Phone, and Message, then save it to a Google sheet.
I can see how it can be useful for freelancers and business owners reduce some manual effort.
Quite simple, but it was a solid choice to get me to learn how to plug different tools together.
My Tools
n8n
Ollama
Gmail
Google sheets
The Flow
Why I picked a open source model?
Before building anything, I needed to choose a large language model (LLM).
I asked ChatGPT to help me compare a few options based on my needs: simple use case, free or low cost, fast information parsing.
Why I chose Gemma:
Fully open-source
Didn’t need GPT-4-level reasoning
Fast output
Great entry point to learn without cloud dependencies
What I didn’t know (and had to learn the hard way)
This project exposed a lot of gaps in my understanding:
I didn’t know what was the difference between local vs cloud hosted
I didn’t understand the different nodes eg: HTTP requests, information extractor etc and to fill out the parameters
I explored a lot of options before figuring out the smoothest way to accomplish the task
I was struggling with setting up the Google credentials for the longest time
I was constantly asking: What do I do now? What does this error mean?
But it’s normal and okay. This was the point. Learning through building.
Useful prompts I used along the way
These prompts helped me figure out how to build it at every step:
"Explain this like I’m 10 years old"
"What’s the first step to implement this?"
"How do I deal with this error?"
"Are there other ways to do this in n8n?"
A easier way would’ve been to find an existing n8n custom GPT or setting up a Gemini Gem/Claude Project/Notebooklm and attach the n8n docs for reference.
Extra tip: If you are using the free versions of AI platforms, you can keep a summary on what you are trying to build so that you can feed context and debug faster when run out of interactions.
If I started again, I would…
Test your prompts using free models before hooking up a paid one
Try a simpler version first eg: extract 1 field
Prepare testing material in advance
Don’t worry about "doing it the right way" → just ship something and learn
What’s happened next
Building this automation from stratch helped me connect the dots between theory and actual usage. Now, I’ve made multiple automations adding more complexity to the systems over time.
Will be sharing more on my AI journey as an creator, educator and consultant in the space.
If you want the n8n template, subscribe and comment below!
And if you're trying to build your first AI automation?
You got this. It's confusing, messy, but worth learning.
Thanks for reading!
Stay inspired,
Viv