rag-api

Introduction

This repository hosts a FastAPI-based API for Retrieval-Augmented Generation (RAG), a cutting-edge technique in Natural Language Processing that enhances response quality by retrieving relevant information during the generation process. Our implementation focuses on speed, scalability, and ease of integration utilizes Milvus for efficient vector storage.

Features

• Fast response times suitable for real-time applications.
• Scalable architecture, handling concurrent requests efficiently.
• Customizable retrieval component for domain-specific applications.

Default Configuration

• Embedding Model: intfloat/multilingual-e5-small (Embedding dimensions: 384)
• Indexing algorithm: HNSW (maxConnections: 32, efConstruction: 128, ef: 64)
• Similarity metric: Cosine distance

Note: The API design is highly flexible, allowing for straightforward substitutions of the embedding model and indexing algorithm. Users can easily replace 'intfloat/multilingual-e5-small' with their preferred model and switch out the HNSW indexing algorithm in Milvus, adapting the system to diverse requirements and use cases.

Supported file types

• pdf
• txt
• docx / doc
• html
• md
• py
• js
• json
• csv
• xml
• xlsx / xls

Install

bash setup.sh

Run

bash setup.sh