mirror of
https://github.com/hanzoai/papers.git
synced 2026-11-02 10:23:36 +00:00
zeekay
f9db4f5737
Bylines, paper authorship, correspondence addresses and the founder blurb under each post named an individual. They name the organisation that published the work instead: the brand on posts and pages, the research byline on papers, and research@ in place of a personal address. Both are what these repositories already used elsewhere — nothing new was coined. Co-authors are untouched. Where a paper named someone else alongside, they stay named; only the one byline changed. Bibliography entries, running heads, PDF metadata and affiliations move with it, so a citation still resolves to whoever the cited paper names. Sample and fixture data uses the placeholder name these tests already use elsewhere, so no assertion changed meaning.
58 lines
2.8 KiB
TeX
58 lines
2.8 KiB
TeX
% Hanzo Search: AI-Powered Search with RAG
|
|
\documentclass[11pt,twocolumn]{article}
|
|
\usepackage{../shared/hanzocover}
|
|
\usepackage{amsmath, amssymb, amsthm}
|
|
\usepackage{mathtools}
|
|
\usepackage{bm}
|
|
\usepackage{graphicx}
|
|
\usepackage{booktabs}
|
|
\usepackage{hyperref}
|
|
\usepackage{enumitem}
|
|
\usepackage{algorithm}
|
|
\usepackage{algpseudocode}
|
|
\usepackage{xcolor}
|
|
\usepackage{float}
|
|
\usepackage{subcaption}
|
|
\usepackage{multirow}
|
|
\usepackage{listings}
|
|
\input{../shared/lstlang}
|
|
\usepackage[numbers]{natbib}
|
|
\hypersetup{colorlinks=true,linkcolor=black,citecolor=blue,urlcolor=blue}
|
|
|
|
\newtheorem{theorem}{Theorem}
|
|
\newtheorem{lemma}[theorem]{Lemma}
|
|
\newtheorem{definition}{Definition}
|
|
\newtheorem{proposition}[theorem]{Proposition}
|
|
|
|
\title{Hanzo Search: AI-Powered Search with Retrieval-Augmented Generation}
|
|
\author{
|
|
Marcus Chen \quad
|
|
David Wei \quad
|
|
Hanzo AI Research \\
|
|
\textit{Hanzo AI Research} \\
|
|
\texttt{[email protected]}
|
|
}
|
|
\date{September 2025}
|
|
|
|
\begin{document}
|
|
|
|
\hanzocoverpage
|
|
|
|
\begin{abstract}
|
|
We present \textbf{Hanzo Search}, an AI-powered search engine that combines retrieval-augmented generation (RAG) with generative user interfaces to deliver synthesized, source-grounded answers instead of traditional ranked document lists. Hanzo Search introduces three key innovations: (i) a \emph{hybrid retrieval pipeline} that fuses sparse (BM25) and dense (neural embedding) retrieval with learned fusion weights, achieving 18.7\% higher NDCG@10 than either method alone on the BEIR benchmark; (ii) a \emph{generative UI framework} that dynamically renders structured answer components (tables, charts, comparisons, timelines, code blocks) based on query intent classification, increasing user engagement by 34\% compared to text-only answers; and (iii) a \emph{source attribution system} that provides sentence-level provenance for every claim in the generated response, achieving 94.3\% attribution accuracy on our SourceVerify benchmark. We evaluate Hanzo Search on standard information retrieval benchmarks (MS MARCO, BEIR, Natural Questions) and a novel generative search benchmark (GenSearch-500), demonstrating state-of-the-art performance in answer quality, factual accuracy, and user satisfaction. Production deployment analysis from 12 months of operation with 2.1 million queries reveals that generative search reduces the average number of follow-up queries by 47\% and increases task completion rates by 28\% compared to traditional search interfaces.
|
|
\end{abstract}
|
|
|
|
\input{sections/01-introduction}
|
|
\input{sections/02-system-architecture}
|
|
\input{sections/03-query-understanding}
|
|
\input{sections/04-hybrid-retrieval}
|
|
\input{sections/05-grounded-generation}
|
|
\input{sections/06-generative-ui}
|
|
\input{sections/07-evaluation}
|
|
\input{sections/08-production-deployment}
|
|
\input{sections/09-related-work}
|
|
\input{sections/10-discussion}
|
|
\input{sections/99-conclusion}
|
|
|
|
\end{document}
|