KNIME 101 - Live Demo - Simple RAG

Improve your RAG pipeline:

• Implement topic based document chunking versus the character splitting method

• Use an LLM component to generate key-word heavy summaries of each document

• Use an LLM to find the best document for the user question and apply this document ID as a metadata filter in the retriever node.

• Use an LLM to rewrite the user query to be more inline with the document content.

• Use an LLM to filter retrieved documents for relevance (latency trade-off here) OR apply a threshold filter based on dissimilarity score (lower is better) to pass the most relevant documents to the model

• Use a more powerful embedding model such as emb-3-large to better capture semantic relationships