I watched the tutorial from planetnocode about using carbon.ai and Claude to build a RAG (retrieval augmented generation) powered AI chatbot - i.e. a chatbot that could add user generated information as an extra layer to it’s thought processes before delivering a response.
This is interesting, but I’m much more interested in creating a RAG that allows users to upload their own documents to be included (the tutorial video only allowed typed text input). Has anyone built anything like this? The obvious use case is as an internal policy and process chatbot for an organisation. There are lots of tutorials on youtube for doing this using agents, and you can use MS copilot studio to create an app in office to do it, but I can’t find anything for bubble.io.
I believe there are several Bubble templates that do something similar. They create OpenAI Assistants and allow for uploading documents. I would suggest checking out the template page. Maybe that would start you in the right direction.
Hi J805, thanks for your advice. I had a look at the template page under AI templates, and I couldn’t find anything that uses retrieval augmented generation, just lots of chatbots linked to different LLMs. It would be great if you could suggest a specific template.
I try to stay away from recommending specific templates because you never know how they are built. I see some that say you can upload documents, are you just looking to pass along user generated information to the AI? That would just be something you could add in the workflow. Unless I am missing something?
What I am looking to set up is a chatbot based on the user’s own uploaded documents, like Notebooklm but integrated into bubble. You can also do this internally in MS Office 265 using copilot studio, which is probably what I’ll do for paid clients, but I want to create a playground where I can demo the concept.
I want to use Retrieval Augmented Generation to do this because based on the code-based tutorials I’ve seen, this is the best technology for doing this.
This is doable using just the Bubble database and a plugin. It can work in the same way as how Pinecone perform the similarity search using dense vectors.
