Metadata-Version: 2.4
Name: rag-eval-kit-avan
Version: 0.1.0
Summary: A lightweight Python wrapper for evaluating RAG pipelines using Ragas.
Author-email: Avan Kumar J <avankumar2025@gmail.com>
License: MIT
Project-URL: Homepage, https://github.com/avankumar2001/eval-rag-system
Project-URL: Repository, https://github.com/avankumar2001/eval-rag-system
Keywords: rag,llm,evaluation,ragas,langchain
Requires-Python: >=3.9
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: ragas
Requires-Dist: datasets
Requires-Dist: langchain
Dynamic: license-file

# rag-eval-kit-avan

A lightweight Python wrapper for evaluating RAG (Retrieval-Augmented Generation) pipelines using [Ragas](https://github.com/explodinggradients/ragas).

If you've built a RAG system with LangChain, LlamaIndex, or plain Python, you know that "it works" is not enough. You need to measure **Faithfulness**, **Answer Relevancy**, **Context Precision**, and **Context Recall**. This package wraps those Ragas metrics into a single function call.

## Installation

```bash
pip install rag-eval-kit-avan
```

## Usage

```python
from rag_eval_kit import evaluate_rag

questions = ["What is RAG?", "What is a vector database?"]
answers = [
    "RAG stands for Retrieval-Augmented Generation.",
    "A vector database stores embeddings for similarity search.",
]
contexts = [
    ["RAG is a technique that combines retrieval with generation..."],
    ["Vector databases like FAISS and ChromaDB store high-dimensional vectors..."],
]
ground_truths = [
    "RAG stands for Retrieval-Augmented Generation.",
    "A vector database stores embeddings and supports similarity search.",
]

results = evaluate_rag(questions, answers, contexts, ground_truths)
print(results)
```

## Why This Exists

Most RAG tutorials stop at "here's how to build a chain." Nobody tells you how to **prove** your RAG pipeline is actually working. This package makes it a one-liner.

Built as part of an [Eval-First RAG System](https://eval-rag-system.streamlit.app) project.

## License

MIT
