Indexing

#context = ModelContext()
#context.set_default()
#store = DocumentStore()

For testing I would start exploring by having a document I want to be able to retrieve information from.

My naive implementation would be an index for the embedding and mapping with the node index. Lets try that.

Todo:

• Try out different scoring strategies weighting other type of things, like metadata similarity, options that have been picked by an LLM and such.
• Docker container to rapidly deploy agents.
• Support own models apart from sentence transformer models.

#Won't be used for now but serves for dependency injection for the index, to try diff retrieval strategies and combine them.
#Will separate the retrieval strategies in the future.  
class Retriever(ABC):
    @abstractmethod
    def retrieve(self, query_embedding, embeddings, top_k):
        pass

---

source

VectorNodesIndex

 VectorNodesIndex (context=None)

Inside here the embeddings stored are normalized. So when doing operations with vectors has to be kept into account.

	Type	Default	Details
context	NoneType	None	May not be needed in postgres.