This workflow implements a Retrieval Augmented Generation (RAG) chatbot that answers employee questions based on company documents stored in Google Drive. It automatically indexes new or updated documents in a Pinecone vector database, allowing the chatbot to provide accurate and up-to-date information. The workflow uses Google’s Gemini AI for both embeddings and response generation.

How it works

The workflow uses two Google Drive Trigger nodes: one for detecting new files added to a specified Google Drive folder, and another for detecting file updates in that same folder.

Automated Indexing: When a new or updated document is detected

1. The Google Drive node downloads the file.
2. The Default Data Loader node loads the document content.
3. The Recursive Character Text Splitter node breaks the document into smaller text chunks.
4. The Embeddings Google Gemini node generates embeddings for each text chunk using the text-embedding-004 model.
5. The Pinecone Vector Store node indexes the text chunks and their embeddings in a specified Pinecone index.
6. The Chat Trigger node receives user questions through a chat interface. The user’s question is passed to an AI Agent node.
7. The AI Agent node uses a Vector Store Tool node, linked to a Pinecone Vector Store node in query mode, to retrieve relevant text chunks from Pinecone based on the user’s question.
8. The AI Agent sends the retrieved information and the user’s question to the Google Gemini Chat Model (gemini-pro).
9. The Google Gemini Chat Model generates a comprehensive and informative answer based on the retrieved documents.
10. A Window Buffer Memory node connected to the AI Agent provides short-term memory, allowing for more natural and context-aware conversations.