Skip to main content
ZICQ

Topics RAG in Practice

RAG in Practice

Retrieval-Augmented Generation from principles to engineering: chunking, vector search, reranking, hybrid retrieval, and production evaluation.

· Aug 20, 2026 #RAG#retrieval-augmented generation#vector search#hybrid search#reranking · 30 / 224 articles

News · 2026-09-12

LlamaIndex Launches Just-in-Time Agentic OCR: A New Paradigm for Intelligent OCR Processing

LlamaIndex has introduced a novel approach called Just-in-Time Agentic OCR, which employs a two-pass system for document processing to enhance intelligent agents' understanding of documents in a data room. The first pass uses a lightweight parsing tool, LiteParse, for rapid text extraction, while the second pass leverages a Vision Language Model (VLM) for in-depth analysis of relevant pages. This method significantly reduces processing costs while maintaining the accuracy required for specific t

News · 2026-09-11

LlamaIndex 0.11 Released: Introducing Event-Driven Workflows and Enhanced Observability

LlamaIndex has released version 0.11, introducing significant updates such as an event-driven Workflows architecture, enhanced instrumentation for better monitoring and debugging, an improved Property Graph Index, and a 42% reduction in package size by removing OpenAI as a core dependency. Additionally, the release includes support for Pydantic V2, asynchronous streaming, and a Structured Planning Agent, aiming to improve the efficiency and scalability of building complex generative AI applicati

News · 2026-09-11

LlamaIndex and Azure OpenAI Launch Guide for Building Serverless RAG Applications

LlamaIndex and Azure OpenAI have jointly released a comprehensive guide for building serverless RAG (Retrieval-Augmented Generation) applications. This guide covers the implementation of the RAG architecture using LlamaIndex, data ingestion, index creation, multi-agent system development, and deployment on Microsoft Azure. It provides developers with a complete AI solution for creating intelligent applications that leverage their own business data.

News · 2026-09-11

LlamaIndex Releases Llama-Deploy: Microservice-Based Deployment for Multi-Agent AI Systems

LlamaIndex has announced the release of Llama-Deploy, a solution for deploying and scaling agentic workflows built with LlamaIndex. Leveraging a microservices architecture, Llama-Deploy offers seamless deployment, scalability, flexibility, and fault tolerance, simplifying the transition from development to production for multi-agent AI systems. Key components include a message queue, control plane, and orchestrator, with an async-first design and state management features.

News · 2026-09-11

LlamaIndex Releases Major LlamaParse Update: Integrates Mistral-3B and Mistral-8B Models

LlamaIndex has announced a major update to its core product, LlamaParse, featuring day-0 integration with Mistral's latest models, Mistral-3B and Mistral-8B. This update aims to enhance the computational capabilities for document parsing and agentic applications, while also supporting more complex multimodal RAG (Retrieval-Augmented Generation) pipeline construction. Additionally, LlamaIndex showcased a project on Dynamic Few-Shot Prompting, demonstrating its potential in customer support and te

News · 2026-09-11

LlamaIndex Launches Next-Gen Report Generation Framework: Automating Knowledge Work with AI

LlamaIndex has launched a next-generation report generation framework based on Retrieval-Augmented Generation (RAG) technology. This framework aims to automate the creation of complex documents such as research reports, presentations, and technical documents. It features five core components: structured output definitions, advanced document processing, knowledge base integration, multi-agent workflow architecture, and template processing systems. By automating the entire process from information

News · 2026-09-11

LlamaIndex Launches Next-Gen Report Generation Framework: Revolutionizing AI-Assisted Knowledge Work Beyond Basic RAG

LlamaIndex has launched a next-generation report generation framework based on Retrieval-Augmented Generation (RAG) technology, designed to automate the creation of complex documents such as research reports, presentations, and technical documentation. The framework leverages five core components—structured output definition, advanced document processing, knowledge base integration, multi-agent workflow architecture, and template processing system—to automate the entire process from information

News · 2026-09-11

LlamaIndex Releases OilyRAGs: RAG-Powered AI Mechanic Assistant with 6000% Efficiency Boost

LlamaIndex has launched OilyRAGs, an AI-powered mechanic assistant that leverages Retrieval-Augmented Generation (RAG) technology to expedite customer service, mechanical maintenance, and repetitive operational tasks. The tool features a multimodal hands-free interface, enabling mechanics to decode model numbers, diagnose issues, and query required parts using their phones, achieving up to a 6000% efficiency boost. This innovation optimizes workflows, reduces carbon emissions, and minimizes huma

News · 2026-09-11

NVIDIA Announces NIM Integration with LlamaIndex for Agent-Enhanced AI Query Engines

NVIDIA has announced the integration of its NIM microservices suite with LlamaIndex, enabling the development of agent-enhanced AI query engines. This integration leverages NIM's capabilities for deploying AI models with LlamaIndex's retrieval-augmented generation (RAG) framework, allowing for the intelligent decomposition and routing of complex queries. By breaking down intricate questions into sub-questions and assigning them to appropriate tools or models, the combined solution enhances the d

News · 2026-09-11

LlamaIndex Releases RAG Context Refinement Agent: AI Agent for Optimizing Code Retrieval-Augmented Generation

LlamaIndex has launched the RAG Context Refinement Agent, an AI-powered tool designed to optimize Retrieval-Augmented Generation (RAG) processes for code repositories. By leveraging iterative context evaluation and refinement mechanisms, the agent helps developers retrieve relevant code snippets and generate precise answers more efficiently, thereby enhancing technical support and development workflows.

News · 2026-09-11

LlamaIndex Integrates with Azure AI: A New Ecosystem for Enterprise-Level RAG Applications

LlamaIndex has announced a deep integration with Microsoft Azure AI, launching an end-to-end Retrieval-Augmented Generation (RAG) solution on Azure. This integration encompasses core components, storage and memory optimization, enhanced agentic tools, and rapid deployment templates, aiming to provide enterprises with an efficient, secure, and scalable AI application development platform. By leveraging Azure OpenAI Service, Azure AI Embeddings, and Azure AI Search, LlamaIndex enables users to bui

News · 2026-09-11

LlamaIndex Releases Permissions-Aware SharePoint Retrieval for Enhanced RAG Application Security

LlamaIndex has launched a permissions-aware content retrieval feature for SharePoint, enabling RAG (Retrieval-Augmented Generation) applications to automatically synchronize SharePoint's granular access controls. This feature, powered by LlamaParse, allows users to control document access based on SharePoint permissions, enhancing data security and compliance. It supports automatic synchronization of permission changes, ensuring that document access in RAG applications remains consistent with Sh

News · 2026-09-10

LlamaIndex Launches Agentic Retrieval: A New Paradigm for Intelligent Information Retrieval Beyond Traditional RAG

LlamaIndex has introduced Agentic Retrieval, a novel approach designed to address the limitations of traditional RAG (Retrieval-Augmented Generation) systems. Leveraging the LlamaParse API, this method intelligently routes queries across multiple knowledge bases, delivering more precise contextual information. Agentic Retrieval supports various retrieval modes, including an auto-routed mode that dynamically selects the most appropriate retrieval strategy based on the query content, thereby enhan

News · 2026-09-10

LlamaIndex Discusses MCP vs. Vector Search: The Future of RAG

LlamaIndex has published an in-depth analysis discussing the trade-offs between the emerging Model Context Protocol (MCP) and traditional vector search in Retrieval-Augmented Generation (RAG) systems. While MCP offers real-time access and simplified data pipelines for structured data, it falls short in handling unstructured data, achieving global semantic understanding, and maintaining consistency across heterogeneous data sources. The article concludes that the most effective approach combines

News · 2026-09-10

LlamaIndex Integrates DeepEval: Enhancing RAG Performance Evaluation and AI Application Development

LlamaIndex has announced its integration with DeepEval, providing developers with a robust toolkit for evaluating and optimizing the performance of RAG (Retrieval-Augmented Generation) applications. Leveraging DeepEval's 50+ pre-built metrics, developers can assess LLM models across dimensions like relevance, faithfulness, and contextual precision, using evaluation results to refine prompt templates, model selection, and hyperparameters, thereby enhancing AI application performance and reliabili

News · 2026-09-10

LlamaIndex Releases LlamaParse: AI-Powered Document Automation Platform for Enterprises

LlamaIndex has launched LlamaParse, an AI-powered document automation platform designed to tackle the challenge of unstructured data parsing in enterprises. The platform excels in precise document parsing, table and chart understanding, and cross-document knowledge connection, providing AI agents with reliable, high-quality contextual data. This enables businesses to automate processes ranging from invoice processing to in-depth research, unlocking new efficiencies and capabilities.

News · 2026-09-09

LlamaIndex Benchmarks Filesystem Tools vs. Vector Search: A New Direction for AI Document Agents

LlamaIndex conducted a benchmark comparing their fs-explorer agent, which leverages filesystem tools for information retrieval, against a traditional Retrieval Augmented Generation (RAG) system. The results demonstrated that while RAG was faster, the fs-explorer agent outperformed in terms of correctness and relevance, particularly when handling complex documents. This study highlights the potential of filesystem-based AI agents in document processing and suggests new avenues for integrating fil

News · 2026-09-09

LlamaIndex Redefines Its Mission: Beyond RAG to Agentic Document Processing

Originally launched as a RAG (Retrieval-Augmented Generation) framework, LlamaIndex has evolved its mission to focus on agentic document processing infrastructure. This shift addresses the limitations of traditional OCR tools in handling complex documents and aims to provide a more efficient and accurate solution for agents to process unstructured data. By leveraging OCR, document parsing, and workflow automation, LlamaIndex seeks to become a key player in automating knowledge work.

News · 2026-09-09

arXiv Research: Complementary Query Rewriting Boosts Performance Under Strong RAG Baseline

A new arXiv study examines the impact of query rewriting strategies on Retrieval-Augmented Generation (RAG) performance under a strong baseline. The research shows that while individual rewriting strategies are at best competitive with the baseline, combining multiple strategies yields significant performance gains. For instance, a four-method union improves HIT@10 by 12.5 points on enterprise data. Additionally, the study introduces a confidence-gated router that dynamically decides whether to

News · 2026-09-08

Hugging Face Releases ReMoMask-2: Revolutionizing Text-to-Motion Generation

Hugging Face has released ReMoMask-2, a novel Text-to-Motion (T2M) model that addresses key challenges in existing Retrieval-Augmented Text-to-Motion (RAG-T2M) systems. By rebuilding the retrieval database within the generator's latent space and introducing Hierarchical Bidirectional Momentum (HBM) contrastive learning, Semantic Spatial-Temporal Attention (SSTA), and Topology Structured Masking (TSM), ReMoMask-2 achieves state-of-the-art accuracy and faster inference. Extensive experiments on Hu

Info · 2026-09-08

Clean Web-to-Markdown Released: Accelerating HTML Extraction for LLMs and RAG

Clean Web-to-Markdown is a specialized tool designed to enhance the efficiency of Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG) systems by accelerating HTML extraction from web pages. It addresses the inefficiencies of traditional web parsing methods when dealing with complex HTML structures, streamlining the data extraction process and significantly improving the speed and accuracy of LLMs and RAG systems in processing web content. This tool provides developers with a mo

Info · 2026-09-07

arXiv Research: Repeated Queries Exhaust an LLM's Brand Recommendations but Not Its Sources

A new arXiv study examines the brand recommendation capabilities of language models (LLMs) when faced with repeated identical buying queries. The research reveals that whether these queries exhaust an LLM's recommendations depends on retrieval capabilities. Engines without web search continued to add new brands in 86-92% of cases at run 15, with a median of 15-31 organizations, while the retrieval-enabled engine saturated by run 10 with a median of 8 organizations. The study also highlights that

News · 2026-09-07

VERGE Framework Released: A Novel Approach for Enhanced Extraction of Early-Onset Colorectal Cancer Symptoms in Clinical

A new method named VERGE has been released on arXiv for extracting early-onset colorectal cancer symptoms and family history risk status from free-text clinical notes. VERGE employs retrieval-augmented generation to propose initial labels and evidence, followed by a bounded verification-refinement cycle to ensure textual grounding and clinical validity. This approach significantly reduces false positives and improves detection accuracy, with precision increasing from 0.764 to 0.849 and MCC from

News · 2026-09-05

Prompt-Engineering Framework for AI Teaching Assistants: Enabling Scalable and Adaptive Education Support

A new research paper on arXiv introduces a prompt-engineering framework designed to personalize AI teaching assistants powered by large language models (LLMs) and retrieval-augmented generation (RAG), such as Jill Watson. The framework adapts responses by analyzing six learner-specific dimensions, yielding 96 distinct learner profiles, and uses Bloom's Taxonomy to estimate cognitive complexity. Experiments and a human study with five participants demonstrate that the framework can significantly

News · 2026-09-04

arXiv Introduces R²Adapter: A Lightweight Routing and Rewriting Adapter for Efficient Hybrid RAG Systems

arXiv recently released a new study on hybrid Retrieval-Augmented Generation (RAG) systems, introducing a lightweight plugin called R²Adapter. This adapter dynamically allocates queries between vanilla and graph-based RAG, reducing unnecessary graph retrieval overhead. Extensive experiments on three multi-hop QA benchmarks demonstrate that R²Adapter reduces graph-based RAG usage by up to 59% while maintaining comparable answer accuracy. This research provides a novel approach for building effici

News · 2026-09-04

Prompt-Engineering Framework for Hybrid Micro-Level Personalization in AI Teaching Assistants

A new study published on arXiv introduces a prompt-engineering-based framework for personalizing general-purpose AI teaching assistants like Jill Watson. The framework adapts responses across six learner-specific dimensions, yielding 96 distinct learner profiles, and uses Bloom's Taxonomy to assess cognitive complexity during interactions. Experimental results demonstrate perceived differences in response style and structure, with statistical analyses identifying learner attributes associated wi

News · 2026-09-03

PRO-STEP: A Novel Step-level Process Reward Optimization Framework for Retrieval-Augmented Generation

PRO-STEP is a novel optimization framework designed to address the error propagation issue in multi-hop reasoning within Retrieval-Augmented Generation (RAG) models. Unlike traditional methods that only optimize the final answer, PRO-STEP introduces step-level supervision by training a generative Process Reward Model (PRM) to evaluate both logical validity and evidential grounding at each step. It employs PRM-guided value tree search to construct preference pairs and optimizes the policy via ste

News · 2026-09-02

AWS Launches Generative AI-Powered Support Operations Platform: Automating SOP Creation and Intelligent Ticket Resolutio

AWS has launched a generative AI-powered support operations platform that automates the creation of Standard Operating Procedures (SOPs) from training videos, provides intelligent ticket resolution through Retrieval-Augmented Generation (RAG), and predicts SLA risks using machine learning (ML). The platform, built on Amazon Bedrock and AWS Strands Agents SDK, enhances operational efficiency by converting unstructured data into actionable insights, enabling real-time guidance for analysts and pro

News · 2026-09-01

MA-RAG Framework Released: Multi-Agent Retrieval-Augmented Generation for Accurate Summarization of Parkinson's Disease

MA-RAG is a novel multi-agent retrieval-augmented generation framework designed to address the limitations of traditional large language models in handling structured longitudinal clinical assessment data for Parkinson's disease. By decomposing clinical reasoning into domain-specialized agents, combining structured fact extraction, and synthesizing clinically grounded summaries through a final verification stage, MA-RAG significantly improves factual correctness and temporal consistency. Experim

Info · 2026-08-31

Dzen Releases Local Embedder: Reduce RAG Costs to 0.24% of OpenAI's Price

Dzen has released a local embedder for retrieval-augmented generation (RAG) tasks, costing only 0.24% of comparable OpenAI services. This solution aims to reduce reliance on cloud-based AI services by enabling on-premises processing, enhancing data privacy and computational efficiency while significantly cutting costs. The innovation provides developers with a more economical and efficient RAG implementation, particularly beneficial for applications with stringent cost constraints and high priva