DataScienceUIBK/Rankify

🔥 Rankify: A Comprehensive Python Toolkit for Retrieval, Re-Ranking, and Retrieval-Augmented Generation 🔥. Our toolkit integrates 40 pre-retrieved benchmark datasets and supports 7+ retrieval techniques, 24+ state-of-the-art Reranking models, and multiple RAG methods.

What it solves

Rankify provides a unified, modular Python toolkit for building and benchmarking retrieval, re-ranking, and Retrieval-Augmented Generation (RAG) pipelines. It simplifies the process of experimenting with different combinations of retrievers, re-rankers, and generators, reducing the friction of integrating multiple state-of-the-art models and datasets.

How it works

Rankify uses a modular architecture that allows users to swap components easily. It offers a high-level "Pipeline API" for quick deployment (e.g., pipeline("rag")) and a more granular workflow for advanced users. The toolkit integrates various retrieval techniques (including BM25, BGE, and ColBERT), 24+ re-ranking models, and multiple RAG generators. It also includes a RankifyAgent for AI-powered model selection and a built-in server to deploy these pipelines as REST APIs.

Who it’s for

Researchers and practitioners in the field of information retrieval and RAG who need a standardized way to benchmark different models, experiment with retrieval pipelines, or deploy search and generation systems into production.

Highlights

  • One-Line Pipeline API: Rapidly create search, rerank, or RAG pipelines with minimal code.
  • Extensive Model Support: Integrates 7 retrieval techniques and 24+ state-of-the-art re-ranking models.
  • AI-Powered Selection: Includes a conversational agent to recommend the best models based on specific use cases.
  • Benchmark Datasets: Provides access to 40+ pre-retrieved benchmark datasets via Hugging Face.
  • Production Ready: Includes a built-in server for REST API deployment and integrations with LangChain and LlamaIndex.
  • Interactive UI: Features a Gradio-based web playground for testing and comparing models.