RAG-chatbot and my implementation of it

Since last time, we’ve had time to learn more about RAG chatbots and how they work. Using dify.ai we have built our own RAG chatbot, and we have learned how to implement one on our portfolio site.

A key part of making a RAG chatbot useful is giving it the right knowledge to draw from — this is where the “retrieval” part of RAG (Retrieval-Augmented Generation) comes in. Instead of relying solely on what the language model already knows, you feed it a curated set of documents it can search through when answering questions.

For my portfolio chatbot, I set up a pipeline that keeps this knowledge base automatically up to date. My about page and all my blog posts are written as .md files in my repository, and every time I push new changes to GitHub, a GitHub Action runs and syncs the latest content over to my dify.ai chatbot. This means the chatbot always has access to my most recent posts and information — without me having to manually update anything.

It was a great way to see RAG in a real, practical context: the chatbot can answer questions about me and my projects by actually retrieving content I wrote, rather than hallucinating or giving generic responses.

During the implementation I also had the chance to get comfortable with making changes to the portfolio, and put my own mark on the template that I use.

Emil Kriegel