RAGWire and ViewRAG
These are complements: RAGWire provides the ingestion and hybrid search infrastructure for multi-format documents, while ViewRAG specializes in layout-aware chunking and multimodal understanding of PDFs, making it a potential upstream processor that could feed structured document knowledge into RAGWire's pipeline.
About RAGWire
laxmimerit/RAGWire
Production-grade RAG toolkit — ingest PDFs, DOCX, XLSX into Qdrant with LLM metadata extraction, hybrid search, and SHA256 deduplication.
Supports multi-format ingestion via MarkItDown (PPTX, XLSX, DOCX, PDFs), markdown-aware recursive chunking, and customizable LLM-based metadata extraction through YAML configuration. Provides pluggable embedding providers (Ollama, OpenAI, HuggingFace, Google, FastEmbed) with Qdrant's dense/sparse/hybrid search, plus directory-level ingestion with file and chunk-level SHA256 deduplication. Designed as modular Python components with environment variable substitution for production deployments.
About ViewRAG
David-Lolly/ViewRAG
图文并茂的 PDF RAG 系统：支持版式感知分块、图表深度理解与精准视觉溯源。 Multimodal PDF RAG: Features layout-aware chunking, visual chart understanding, and precise inline image citations.
Implements a multimodal RAG pipeline combining PaddleX layout-aware PDF parsing with vision LLM understanding of charts and images, storing structured semantic descriptions in pgvector for retrieval. The system uses OpenAI-compatible APIs for flexible model selection (Qwen, DeepSeek, GLM, Ollama) and integrates MinIO for image storage, enabling inline image citations in LLM responses with precise PDF page/section tracing through a custom reference attribution system.
Related comparisons
Scores updated daily from GitHub, PyPI, and npm data. How scores work