GenAI RAG and Bonus

[ahmet-cskn]

The GenAI chatbot now incorporates the notes, calendar events, and checklists data of the user in its responses. I also implemented the optional GenAI bonus feature (RAG architecture with vector database). Below is the application flow:
-The user sends a prompt to the chat endpoint.
-The data of the user from the notes, calendar, and checklist services are fetched using the “get all” endpoints of the respective services.
-The data is uploaded to the Weaviate vector database (which runs locally as a separate docker container). Each data item gets vectorized.
-A vector similarity check is ran (using a local model) between each data item in the database and the prompt itself, in order to determine which data items are relevant to the user’s prompt.
-The relevant data items get included in the prompt to establish context, then the prompt is routed to the LLM, and the chat endpoint sends the LLM response as usual.

Other changes:
-For the GenAI service to be able to retrieve data from other services, there should be a network between them. That’s why “internal-net” was added to the docker compose
-I hardcoded some actual data to the stub endpoints in the notes, calendar, checklist services, to be able to test this feature

[alexander-wudy]

Topic 1

-A vector similarity check is ran (using a local model) between each data item in the database and the prompt itself, in order to determine which data items are relevant to the user’s prompt.

"using a local model". Where exactly in the code is this model running? And how is it running, i.e. via an additional container or do we have to install something locally?

Topic 2

Is it possible that you can add some tests to your code, also as a whole? It would also be alright to do it in an addtional issue, but as of right now your code is not tested anywhere iirc, right?

[ahmet-cskn]

1. No new container or any manual installation is needed. It is a small model that is downloaded to the GenAI component when the app starts running.
2. I did test it myself locally using curl commands but you are correct that the code itself doesn’t have tests. I’ll definitely add them👍🏻