Introduction
Recently, I have been exploring various features of n8n that I just started to use. I found that when creating AI processes, if one of the process does not meet expectations, restarting the entire process can be very time-consuming and costly. For example, if you are creating a story generator with different AI generations: character settings, story outlines, dialogue content, etc., if any of these do not meet expectations, the entire process has to be scrapped and modified. At this point, we can add human feedback, modifying prompts repeatedly until we are satisfied.
I also utilized this feature during my article research with generated AI to dynamically adjust the needs in communication with AI.
Human In the Loop
HIL is a feature introduced in n8n 1.78, which can be selected in the node creation field, with options for communication software like Discord, Slack, Telegram, etc., allowing feedback to be given anytime and anywhere.
Its function is essentially to send a response and wait, with several expected response formats:
- Approval
- FreeText
- CustomForm
The purpose is to select the corresponding communication software integration and send the corresponding action links and messages, enabling people to provide feedback.
The node will automatically wait and return the submitted results. For example, I can send a piece of data to the AI node like this, and after asking for human opinions, if a regeneration is needed (analyzing feedback for further action, very cool isn’t it? 😆), create a new feedback entry, inputting the latest generated article and feedback back to AI for refinement:
[ { "feedback": "...", "article": "..." }]