Core / RAG Workflow Engine

This section dives into the engine of the stack—the part that powers your Retrieval-Augmented Generation (RAG) solutions with just a YAML configuration file.

Core Workflow

The diagram below illustrates how the Core workflow processes user queries and documents step by step:

Here’s what happens under the hood:

Step 1: Init

The process begins by loading your YAML configuration.

Step 2: Parse YAML Config

The configuration file is parsed, serving as the blueprint for the entire workflow.

Step 3: Initialize Tools

All the tools defined in the configuration are loaded and prepared for action.

Step 4: Create or Retrieve Collections

If working with a vector database, the system creates a new collection or retrieves an existing one for efficient data handling.

Step 5: Process and Store Documents

Documents are processed, chunked, and stored in the vector database within the appropriate collection. Each RAG tool manages its own dedicated collection.

Step 6: Classify and Select Tools (Documents)

Incoming documents are classified, and the right tool is selected for processing and storing the partitioned data chunks.

Step 7: Handle User Queries

The AI agent interacts with users, leveraging tools and the database to provide insights and responses.

Step 8: Classify and Select Tools (Queries)

Incoming queries are classified, and the most suitable tool is chosen. For RAG tools, this may involve retrieving stored documents. For prompt tools, a direct answer is generated.

Step 9: Retrieve Relevant Chunks

When required, relevant data chunks are fetched from the vector database to enhance responses.

---