https://github.com/emredeveloper/rag-with-cache

A comprehensive study and implementation of best practices for Retrieval-Augmented Generation (RAG) systems. This project includes advanced features like model caching, multilingual support, and evaluation metrics to enhance the performance and usability of RAG systems.
https://github.com/emredeveloper/rag-with-cache

artificial-intelligence documentation llm python rag

Last synced: about 1 month ago
JSON representation

A comprehensive study and implementation of best practices for Retrieval-Augmented Generation (RAG) systems. This project includes advanced features like model caching, multilingual support, and evaluation metrics to enhance the performance and usability of RAG systems.

• Host: GitHub
• URL: https://github.com/emredeveloper/rag-with-cache
• Owner: emredeveloper
• Created: 2025-01-20T21:21:21.000Z (over 1 year ago)
• Default Branch: main
• Last Pushed: 2025-10-03T15:48:45.000Z (9 months ago)
• Last Synced: 2025-10-13T22:43:49.752Z (8 months ago)
• Topics: artificial-intelligence, documentation, llm, python, rag
• Language: Python
• Homepage:
• Size: 15.7 MB
• Stars: 1
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • Security: SECURITY.md

Awesome Lists containing this project

README

          
# 🚀 RAG-With-Cache

A compact reference implementation of **Retrieval-Augmented Generation (RAG)**

that focuses on the core building blocks: PDF ingestion, vector search, and

language model generation. The codebase has been simplified so the main RAG

pipeline lives in a dedicated `rag/` package with minimal dependencies and a

clean entry point.

---

## 🌟 Features

- **Modular package layout** – Embedding, language, data-loading, and

  retriever utilities are organised under `rag/` for easier reuse.

- **FAISS and HyDE retrievers** – Switch between traditional dense retrieval

  and HyDE-style hypothetical document retrieval from the command line.

- **Model caching** – Embedding and language models are cached locally to

  avoid repeated downloads.

- **PDF utilities** – Lightweight helpers for loading and chunking documents.

- **Test coverage** – Pytest suite that exercises each major component with

  fast stubs.

---

## 🚀 Quick Start

1. **Install dependencies**

   ```bash

   pip install -r requirements.txt

   ```

2. **Prepare your documents**

   Place PDFs inside `data/pdfs/`. The directory is created automatically when

   running the CLI, but adding files ahead of time lets you test retrieval

   immediately.

3. **Run the demo CLI**

   ```bash

   python main.py

   ```

   Choose between FAISS or HyDE retrieval when prompted and start asking

   questions about your documents.

---

## 📁 Project Structure

```text

├── rag/

│   ├── config.py             # Configuration dataclass

│   ├── data/                 # PDF loading utilities

│   ├── embeddings.py         # SentenceTransformer wrapper with caching

│   ├── language.py           # HuggingFace causal LM wrapper with caching

│   ├── retrievers/           # FAISS and HyDE retrievers

│   └── system.py             # High-level RAG orchestration

├── data/pdfs/                # PDF documents for retrieval

├── tests/                    # Pytest suite

├── main.py                   # Command-line entry point

├── requirements.txt          # Minimal dependency set

└── setup.py                  # Package metadata

```

---

## 🧪 Running the Tests

```bash

pytest

```

The HyDE tests rely on stubbed models so they run quickly without downloading

large checkpoints.