I created a graph storage from dozens of annual stories (with tables)
Changing PDFs to textual content was attainable however has by no means been simpler.
I lately created a graph knowledge retailer for use in an RAG. In different phrases, we constructed a GraphRAG.
Graph RAGs are a implausible different to different RAG apps like broadly used vector store-backed RAGs. They create reasoning to the desk. For instance, with semantic similarity search (the method utilized in vector shops to retrieve data), you might ask who the CFO of XYZ, Inc. was final yr. As a result of XYZ, Inc.’s final yr’s annual report would explicitly point out its CFO. However consider a query like this: Which two administrators of XYZ, inc. have studied in the identical college? The retrieval course of received’t be capable of fetch the related data with out mentioning a college identify. However graph RAG may do it.
Nonetheless, the important thing situation right here is how we assemble the graph for retrieval. I’ve addressed this situation in a separate publish lately. Pondering one other step backward, how will we even put together the annual…