OmniRAG

3 applications chosen by Henry Pereira

Three composable gateways that together form a complete RAG pipeline on InterSystems IRIS: omniIngest parses and chunks documents, omniEmbedding turns text into vectors across 10 providers, and omniReRank reorders retrieval results for relevance — each usable standalone, or chained end to end.

First published 04 Nov, 2026Last updated 04 Nov, 2026

omniEmbedding

The foundation: a pure ObjectScript gateway that centralizes embedding generation across 10 providers (Ollama, OpenAI, Azure, Cohere, Gemini, Mistral, Voyage, Jina, Bedrock, VertexAI) behind one %Embedding.Config-driven interface — so the rest of the pipeline never hardcodes a provider.

omniRerank

The precision layer: reranks whatever omniEmbedding (or any vector search) retrieves, across Cohere, Voyage, Jina and Ollama, so the final context sent to the LLM is ordered by actual relevance, not just raw cosine similarity.

omniIngest

The entry point: a PyProd-based Interoperability Production that turns raw documents (PDF, DOCX, PPTX, MD, TXT) into structurally-aware chunks via Docling, then hands them to omniEmbedding — closing the loop from file to vector without any manual preprocessing step.

This bundle is community-generated content and is not supported by InterSystems.