Source Code Overview

The src/nvidia_rag directory contains the core implementation of the project. Below is an updated overview of the subdirectories and files, along with their purposes:

Directory Structure

ingestor_server/

This directory contains the implementation for the ingestion server, which handles document ingestion and related tasks.

• __init__.py: Initializes the module.
• Dockerfile: Docker configuration for the ingestion server.
• main.py: Entry point for the ingestion server.
• nvingest.py: Handles NVIDIA-specific ingestion logic.
• server.py: Defines the FastAPI server for handling ingestion-related APIs.
• task_handler.py: Manages ingestion tasks, including task submission and handling.

rag_server/

This directory contains the implementation of the RAG (Retrieval-Augmented Generation) server.

• __init__.py: Initializes the module.
• Dockerfile: Docker configuration for the RAG server.
• health.py: Implements health check APIs for the server.
• main.py: Entry point for the RAG server.
• prompt.yaml: Configuration for prompts used in RAG workflows.
• reflection.py: Implements reflection logic for context relevance and response groundedness.
• response_generator.py: Handles response generation logic.
• server.py: Defines the FastAPI server for RAG-related APIs.
• tracing.py: Enables tracing and instrumentation for the RAG server.
• validation.py: Implements validation logic for RAG responses.
• vlm.py: Implements Vision Language Model integration.

utils/

This directory contains utility modules used across the project.

• __init__.py: Initializes the module.
• common.py: Provides common utility functions.
• configuration.py: Handles configuration management.
• configuration_wizard.py: Provides a wizard for setting up configurations.
• embedding.py: Implements embedding-related utilities.
• llm.py: Contains utilities for working with large language models.
• minio_operator.py: Provides utilities for interacting with MinIO object storage.
• reranker.py: Implements reranking logic for retrieved documents.
• vectorstore.py: Manages vector storage for embeddings.

observability/

This directory provides tools for monitoring and instrumentation.

• langchain_callback_handler.py: Implements callback handlers for Langchain SDK.
• langchain_instrumentor.py: Provides OpenTelemetry instrumentation for Langchain.
• otel_metrics.py: Defines metrics collection using OpenTelemetry.

__init__.py

This file initializes the src module, making it importable as a package.