accuralai-rag

High-performance Retrieval Augmented Generation (RAG) utilities built for the AccuralAI orchestration ecosystem. The package provides:

• Multi-vector embedding utilities with optional GPU + hybrid sparse support.
• Intelligent chunking and metadata extraction helpers tuned for large documents.
• Hybrid dense/sparse retrieval with reciprocal rank fusion and reranking hooks.
• Query optimizers (HyDE, decomposition, semantic augmentation) plus advanced context builders.
• UltraFastRAG pipeline that combines the components with aggressive caching and streaming-friendly hooks.

All heavy dependencies are optional. Install extras such as torch, transformers, faiss, or redis based on your production topology:

pip install -e packages/accuralai-rag[torch,transformers,faiss,redis]

See accuralai_rag/pipeline.py for the orchestrated entry point and tests/ for lightweight examples.