I recently started working with n8n again, this time as a workflow automation engineer. My role involves automating processes to free up teams such as Customer Support, Sales and Finance from the more menial tasks associated with their day-to-day jobs. This involves tasks such as filling out the same forms, applying the right categories and tags, or clicking the same buttons repeatedly every day. Like virtually all modern businesses, we want to leverage AI wherever possible and useful.
This of course means that I need to be familiar with the LLMs available. As a basic introduction to LLMs, I have built a workflow that essentially hypes me up for my first ever trip to Japan.
Every day, it sends me a quick chat message through Telegram, recommending a new place to visit during my trip and sharing a photo of it. The workflow combines LLMs with traditional API calls utilising Bing's REST API to search for a place's website and a photo. It also keeps track of places it has previously suggested in a Notion database, helping to avoid repetition:
I had previously extracted the website information from the LLM, but virtually all the models I tested made mistakes (the hostname was usually correct, but the path was wrong half of the time). The resulting messages look like so:
For example, the workflow can easily be modified to use a different data storage other than Notion, to send messages via a different messenger or channel, or to use a different AI model (including local ones). If you feel like treating yourself to some LLM-generated travel tips as well, you can grab the workflow JSON over on GitHub:
262588213843476