RAG Factory

Project Overview

RAG Factory is an open-source AI platform that transforms virtually any data source into a searchable knowledge base for Large Language Models. The project provides a unified ingestion, indexing, retrieval, and generation pipeline capable of processing documents, websites, repositories, cloud storage, and structured data sources.

Unlike traditional RAG implementations that focus on a single document format or vector database, RAG Factory was designed as a universal retrieval platform with pluggable components for ingestion, embeddings, vector stores, and LLM providers.

The goal is to simplify enterprise knowledge management by providing a consistent framework for converting fragmented information into AI-accessible context while remaining independent of any specific model vendor or storage technology.

---

Key Features

Universal Data Ingestion

Supports ingestion from multiple data sources including:

• Text documents
• PDF and Office documents
• OpenDocument formats (ODF)
• Websites and URLs
• Git repositories
• AWS S3 buckets
• Markdown documentation
• Structured datasets

Pluggable Vector Storage

Supports multiple vector database backends through a common abstraction layer:

• PostgreSQL + pgvector
• ChromaDB
• Future extensibility for additional vector stores

Multi-Provider LLM Support

Provides a unified interface for interacting with multiple model providers:

• OpenAI
• Mistral AI
• Ollama
• Self-hosted models

Retrieval-Augmented Generation

Implements semantic search, retrieval, context assembly, and response generation workflows for knowledge-intensive applications.

End-to-End User Experience

Includes a React-based frontend and FastAPI backend for ingestion, indexing, administration, and querying.

---

My Contributions

• Designed the overall architecture and plugin-based framework.
• Built the document ingestion pipeline supporting heterogeneous data sources.
• Implemented chunking, embedding generation, and indexing workflows.
• Developed abstraction layers for vector databases and LLM providers.
• Integrated pgvector and ChromaDB storage backends.
• Implemented retrieval pipelines and semantic search capabilities.
• Developed FastAPI backend services for ingestion and querying.
• Built React-based user interfaces for knowledge management and search.
• Created deployment workflows and open-source project infrastructure.
• Authored documentation, examples, and developer onboarding guides.

---

Technical Highlights

Universal Knowledge Ingestion

Designed a modular ingestion framework capable of normalizing diverse data sources into a consistent retrieval pipeline.

Vendor-Neutral Architecture

Decoupled retrieval, storage, embedding, and generation layers, enabling users to switch providers without application-level changes.

Extensible Plugin Framework

Implemented provider abstractions that allow new vector stores, embedding models, and LLMs to be integrated with minimal effort.

Production-Oriented Design

Focused on scalability, maintainability, and deployment flexibility rather than building a single-purpose chatbot.

Enterprise Knowledge Enablement

Created a foundation for internal copilots, document assistants, knowledge search platforms, and AI-powered workflows.

---

Challenges & Solutions

Challenge

Most RAG frameworks are tightly coupled to specific document formats, vector databases, or model providers, making migrations and enterprise adoption difficult.

Solution

Built a modular architecture that separates ingestion, embedding generation, retrieval, storage, and generation into independent components connected through well-defined interfaces.

Outcome

RAG Factory enables organizations to build retrieval-powered AI applications while remaining flexible in their choice of data sources, vector databases, and language models.

---

Technology Stack

Backend FastAPI, Python

Frontend React, TypeScript

Vector Databases pgvector, ChromaDB

LLMs OpenAI, Mistral, Ollama

Embeddings Sentence Transformers, OpenAI Embeddings

Storage PostgreSQL, AWS S3

AI Retrieval-Augmented Generation (RAG), Semantic Search

Architecture Plugin Framework, Provider Abstractions, Multi-Tenant Knowledge Systems

Domain Enterprise Search, Knowledge Management, AI Infrastructure, Agentic Systems