This study presents the first application of a Retrieval Augmented Generation (RAG) model for Nepali legal question answering, addressing data scarcity in low-resource languages. Using BM25 on chunked documents, the RAG pipeline achieved high precision and truthfulness, demonstrating its effectiveness in the Nepali legal domain.
Lyzr Cognis introduces a unified memory architecture for conversational AI agents, addressing the lack of persistent memory through a multi-stage retrieval pipeline. It combines a dual-store backend, context-aware ingestion, and temporal boosting, achieving state-of-the-art performance on two independent benchmarks.
This paper proposes a companion-specific governance profile for single-user knowledge wikis, addressing the unique failure mode of entrenchment under user-coupled drift. It discusses emerging personal AI memory architectures from 2026, including RAG-based systems and wiki-style designs, alongside established academic and production memory systems.
Local LLMs often provide poor answers from personal knowledge bases, not due to the model itself, but due to issues in the retrieval layer. This article explores the frustrating problem and how the retrieval pipeline works.
Este trabalho propõe uma formulação rigorosa para a recuperação consciente da diversidade em Geração Aumentada por Recuperação (RAG), abordando a falta de garantias teóricas e escalabilidade dos métodos existentes. A solução utiliza programação quadrática binária com restrição de cardinalidade (CCBQP) e um algoritmo baseado em Frank-Wolfe, demonstrando desempenho superior na fronteira de Pareto de relevância-diversidade e maior velocidade.
This paper proposes a unified multi-agent AI architecture to automate end-to-end machine learning (ML) pipeline generation from datasets and natural-language goals. The five-agent system integrates RAG, an explainable hybrid recommender, and an LLM-based self-healing mechanism, achieving an 84.7% success rate and improved robustness.
Part 2 of this series focuses on building the interface and reasoning engine for a RAG-powered customer agent. It details how to connect a messaging front-end (Telegram Bot API) to a vector database (Pinecone) and an LLM using Make.com to provide accurate, real-time responses.
This document describes a prompt engineering technique that improves AI model response quality by replacing static examples with semantically similar ones retrieved from a vector database. It involves indexing successful conversations and injecting the most relevant examples into the system prompt for complex tasks.
This paper introduces VLM-DeflectionBench, a new benchmark for Large Vision-Language Models (LVLMs) focusing on deflection and hallucination when dealing with conflicting or insufficient evidence. It also proposes a dynamic data curation pipeline to maintain benchmark difficulty over time and a fine-grained evaluation protocol to disentangle model behavior.
CogRAG+ is a training-free framework designed to diagnose and remediate memory and reasoning deficiencies in large language models for professional exam QA. It decouples and aligns retrieval and reasoning with human cognitive hierarchies, employing Reinforced Retrieval and cognition-stratified Constrained Reasoning to enhance accuracy and consistency.
AdaGATE is a training-free evidence controller for multi-hop Retrieval-Augmented Generation (RAG) designed to address noisy or redundant retrieved evidence in limited contexts. It frames evidence selection as a token-constrained repair problem, combining entity-centric gap tracking and targeted micro-query generation to balance coverage, corroboration, and novelty.
This paper introduces CanLegalRAGBench, a new Canadian legal QA benchmark for evaluating Retrieval-Augmented Generation (RAG) systems using realistic queries and expert-annotated case law answers. It highlights the sensitivity of retrieval performance, the competitiveness of open-source embedding models, and the limitations of automatic evaluations and LLM hallucinations in generated responses.
This paper evaluates a Generative AI-based virtual assistant utilizing Retrieval-Augmented Generation (RAG) to support Maastricht University students with project regulations. The system aims to address challenges like hallucinations and provide accurate, context-specific responses by integrating domain-specific knowledge.