podman-desktop-extension-ai-lab
podman-desktop-extension-ai-lab copied to clipboard
UX: RAG data entry mockup
UI for users to connect to vector db of their data for RAG usage of model
Notes for 6 Feb program call:
Concept of an embedding model is unique to RAG patterns, may have implications on AI Studio UI.
RAG pattern:
- choose gen AI model AND an embedding model
Erik has a deck for vector dbs, can share:
https://docs.google.com/presentation/d/1FkprZvdfkYiBjmWW5xDgOPc29XymVIIMzuzU0ipaA-A/edit#slide=id.p
Embedding model is the indexing thing... ask question in chat with model. For RAG - that question gets run through this embedding model, puts it into the embedding space.... when you index all your docs, you run them through that same model. It'sl ike a database, the index part is embedding vectors. Instead of doing exact lookup on a key, it's looking for documents with vectors closest to the thing.
@mairin is there any draft mockup i can start looking at? I was updating the catalog but i need to work on this before adding the rag recipe
Hi Luca, nothing yet :-( I spoke to Jeff about this yesterday. After I get the extensions mocks done today its next on my list. ~m
@nichjones1 @mairin @lstocchi shall we reassign the issue? Do we have mock ups?
@vrothberg I've added Emma as an assignee. We will discuss at our next checkin
I had a look in Penpot for mock-ups for this, and I found this board: https://design.penpot.app/#/workspace/76981dd7-ec3b-802f-8001-9a56ef3fbc9d/52961d58-0a92-80c2-8003-2a6919ad37d6?page-id=e6a8dc76-7670-80e5-8003-ccc1d3d7666d
Are these the right screens?
I do not know, but @mairin @nichjones1 or @slemeur may know.
If this board is about RAG applications, the mock ups are about image generation which is a different topic.
Ok. What is the user flow to get to this page? I'm not familiar with RAG applications and the layout, so I'll have to do some research beforehand.
Ok so I had another look at this. I suppose my question would be what fields / information do we need from the user? Is this something that will be on a form page - and if so where within the application does it fit in? Is it going to be in Catalog > Models > Import? Or is it going to have its own menu item to navigate to it? I need more information about this before I can create a mockup.
I woujld add that RAG recipes would need:
- an additional embedding model to start from
We may also add a datastore from which the embeddings will be computed but I'm not sure this may not be handled by the recipe itself
@slemeur @jeffmaury can you please confirm that the earlier comments in this ticket are still valid and elaborate some more on the exact requirements? Who is the user and what are they trying to do through RAG data entry?
@slemeur @jeffmaury can you please confirm that the earlier comments in this ticket are still valid and elaborate some more on the exact requirements? Who is the user and what are they trying to do through RAG data entry?
From my POV yes but need @slemeur confirmation
is this RAG demo relevant here? https://youtu.be/mouu4PR5Dqg?si=eLBieGqH0-ZRgP4L
@jeffmaury can you please help me understand, who is is the user, who wants to connect their vector db and what are they doing? It could be in a from of a user story: As a ..., I want to ..., so that I can .... I.e. is this an additional story to the fine-tuning or is this part of the recipe or is it something else entirely?
It would be helpful to understand, what screen are they starting from and what their goals are when they are bringing the database in.
@jeffmaury can you please help me understand, who is is the user, who wants to connect their vector db and what are they doing? It could be in a from of a user story: As a ..., I want to ..., so that I can .... I.e. is this an additional story to the fine-tuning or is this part of the recipe or is it something else entirely?
It would be helpful to understand, what screen are they starting from and what their goals are when they are bringing the database in.
Yes that's not entirely clear to me but I can say that it's disconnected from the fine tuning workflow For my POV, is is either:
- part of the workflow starting RAG based recipes
- a new workflow to help users configure their RAG based AI application
Maybe @slemeur can help
HTH
Closing, we will revisit this topic and create a new epic