Automated Document Processing and Q&A with Langchain and Mistral AI

Overview

This n8n workflow is designed to ingest local documents, process them into manageable chunks, and then leverage the power of Langchain and Mistral AI for advanced question answering. The workflow starts by monitoring a local directory for new files using the Local File Trigger. It then extracts relevant metadata like project and filename using a Settings node. The Prep Incoming Doc node prepares the document text for further processing. The core of the document processing involves the Default Data Loader to load document content, Recursive Character Text Splitter to break down large documents into smaller, queryable chunks, and Embeddings Mistral Cloud to create vector embeddings for these chunks. These embeddings are stored and managed in a Qdrant Vector Store. When a question is posed, the Vector Store Retriever fetches relevant information from the vector store. Finally, the Mistral Cloud Chat Model nodes are used to generate answers based on the retrieved context, with the Item List Output Parser and Aggregate nodes helping to structure and present the final response. This workflow is particularly useful for building RAG (Retrieval Augmented Generation) systems where you need to query and extract information from a collection of documents.