Traditional vision search follows the multi-stage cascading architecture (MCA) paradigm to balance efficiency and conversion, yet optimization objectives collide across stages, making it difficult to achieve Pareto optimality. Meanwhile, pronounced multi-view discrepancies of the same item undermine retrieval accuracy, as embedding-similarity methods fail to bridge large appearance variations. In this paper, we propose OneVision, an end-to-end generative framework that addresses both challenges. OneVision introduces VRQ, a vision-aligned residual quantization method that assigns consistent shallow-level codes to visually similar images of the same item while preserving fine-grained differences at deeper levels. It further employs a multi-stage training pipeline to retrieve all relevant items across views for a query image and incorporates user-specific signals for personalized preference modeling. In offline evaluations, OneVision performs on par with online MCA. Dynamic pruning further reduces its inference latency by 18% over the unpruned variant. In A/B tests, it achieves significant improvements: +3.61% item CTR, +2.43% CVR, and +5.89% order volume. The code will be publicly available.

Retrieval-Augmented Generation (RAG) has emerged as a promising approach to address key limitations of Large Language Models (LLMs), such as hallucination, outdated knowledge, and lacking reliable reference. However, current RAG frameworks often struggle with identifying whether retrieved documents meaningfully contribute to answer generation. This shortcoming makes it difficult to filter out irrelevant or even misleading content, which notably impacts the final performance. In this paper, we propose Document Information Gain (DIG), a novel metric designed to quantify the contribution of retrieved documents to correct answer generation. DIG measures a document’s value by computing the difference of LLM’s generation confidence with and without the document augmented. Further, we introduce InfoGain-RAG, a framework that leverages DIG scores to train a specialized reranker, which prioritizes each retrieved document from exact distinguishing and accurate sorting perspectives. This approach can effectively filter out irrelevant documents and select the most valuable ones for better answer generation. Extensive experiments across various models and benchmarks demonstrate that InfoGain-RAG can significantly outperform existing approaches, on both single and multiple retrieval paradigm. Specifically on NaturalQA, it achieves the improvements of 17.9%, 4.5%, 12.5% in exact match accuracy against naive RAG, self-reflective RAG and modern ranking-based RAG respectively, and even an average of 15.3% increment on advanced proprietary models GPT-4o across all datasets. These results demonstrate the feasibility of InfoGain-RAG as it can offer a reliable solution for RAG in multiple applications.