3.x QA: RAG Pipeline about prefect HOT 1 CLOSED

I built a RAG pipeline that takes stores some of our documentation in Pinecone (vector database) and uses relevant documents as context when invoking GPT2. Confidence is calculated as how closely documents match. If confidence is low the flow is paused and a human is asked for intervention, the human's intervention is the relayed to the user.

I didn't really find any failures or frictions in this, most of the friction was in the learning side of this task. As far as incorporating this type of project into prefect it worked fine. Pausing/resuming worked as expected, I was able to use flow run inputs to communicate between processes for user interaction, etc.

rag.zip

from prefect.