mirror of
https://github.com/hanzoai/papers.git
synced 2026-10-30 14:58:42 +00:00
zeekay
0d1d121451
Fix shared/ include paths to ../shared/ for 55 subdirectory papers so
that hanzocover.sty and lstlang.tex resolve when building from each
paper's own directory (the build pattern used by latexmk per paper).
Build fixes:
- hanzo-operative/sections/03-screen-understanding-pipeline.tex: move
\COMMENT from after \FOR{...} onto the \STATE line so the algorithmic
env doesn't mis-parse.
- hanzo-network-architecture (abstract + sections 04, 12, 13): escape
unicode mu (U+03BC) as $\mu$ for pdflatex.
- ntt-transform/sections/08-benchmark-results.tex: same mu escape.
All 55 Tier-2/3 papers now build clean with latexmk -pdf.
110 lines
3.8 KiB
TeX
110 lines
3.8 KiB
TeX
% Hanzo AI Cloud: Confidential Inference, Agent Hosting, and Model Serving
|
|
\documentclass[11pt]{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{multirow}
|
|
\usepackage{listings}
|
|
\input{../shared/lstlang}
|
|
\usepackage{tikz}
|
|
\usetikzlibrary{shapes,arrows,positioning,fit}
|
|
\usepackage{pgfplots}
|
|
\pgfplotsset{compat=1.18}
|
|
|
|
\hypersetup{colorlinks=true,linkcolor=black,citecolor=blue,urlcolor=blue}
|
|
|
|
\definecolor{codegreen}{rgb}{0,0.6,0}
|
|
\definecolor{codegray}{rgb}{0.5,0.5,0.5}
|
|
\definecolor{codepurple}{rgb}{0.58,0,0.82}
|
|
\definecolor{backcolour}{rgb}{0.95,0.95,0.92}
|
|
|
|
\lstdefinestyle{mystyle}{
|
|
backgroundcolor=\color{backcolour},
|
|
commentstyle=\color{codegreen},
|
|
keywordstyle=\color{codepurple},
|
|
numberstyle=\tiny\color{codegray},
|
|
stringstyle=\color{codegreen},
|
|
basicstyle=\ttfamily\footnotesize,
|
|
breakatwhitespace=false,
|
|
breaklines=true,
|
|
captionpos=b,
|
|
keepspaces=true,
|
|
numbers=left,
|
|
numbersep=5pt,
|
|
showspaces=false,
|
|
showstringspaces=false,
|
|
showtabs=false,
|
|
tabsize=2
|
|
}
|
|
\lstset{style=mystyle}
|
|
|
|
\newtheorem{theorem}{Theorem}
|
|
\newtheorem{definition}{Definition}
|
|
|
|
\title{HANZO AI CLOUD\\[0.3em]
|
|
{\large Confidential Inference, Agent Hosting, and\\Model Serving on Sovereign Infrastructure}}
|
|
\author{
|
|
Zach Kelling \\
|
|
\textit{Hanzo Industries Inc (Techstars '17)} \\
|
|
\texttt{[email protected]}
|
|
}
|
|
\date{March 2026}
|
|
|
|
\begin{document}
|
|
|
|
\hanzocoverpage
|
|
|
|
% ============================================================================
|
|
% ABSTRACT
|
|
% ============================================================================
|
|
\begin{abstract}
|
|
We present Hanzo AI Cloud, an infrastructure layer for AI workloads where the
|
|
provider never sees the user's data. Current AI cloud platforms (OpenAI,
|
|
Anthropic, Replicate, Together) require users to transmit plaintext prompts,
|
|
receive plaintext completions, and trust the provider not to log, train on, or
|
|
leak their inputs. This trust model is incompatible with regulated industries
|
|
(healthcare, finance, defense) and with any user who values privacy as a
|
|
technical guarantee rather than a policy promise.
|
|
|
|
Hanzo AI Cloud eliminates this trust requirement through three mechanisms:
|
|
(i)~confidential inference via CKKS fully homomorphic encryption, where inputs
|
|
are encrypted client-side, evaluated homomorphically on the provider's GPUs,
|
|
and decrypted only by the client; (ii)~agent hosting with capsule isolation,
|
|
where each agent receives a private vault, a decentralized identifier, a
|
|
capability budget, and encrypted memory that no other agent or the platform
|
|
can access; and (iii)~model serving with verifiable attestation, where every
|
|
inference result is accompanied by a cryptographic receipt anchored to the Lux
|
|
A-Chain, proving which model produced it, on what hardware, at what time.
|
|
|
|
We describe the system architecture, the FHE inference pipeline, the agent
|
|
capsule model, the model registry, Hanzo Candle (the Rust ML framework
|
|
powering GPU inference), the Jin multimodal architecture, edge inference for
|
|
mobile and IoT devices, and the provider marketplace with stake-weighted
|
|
selection and SLA enforcement.
|
|
\end{abstract}
|
|
|
|
% ============================================================================
|
|
|
|
\input{sections/01-introduction}
|
|
\input{sections/02-the-ai-cloud-problem}
|
|
\input{sections/03-confidential-inference}
|
|
\input{sections/04-agent-hosting}
|
|
\input{sections/05-model-serving}
|
|
\input{sections/06-hanzo-candle}
|
|
\input{sections/07-jin-architecture}
|
|
\input{sections/08-edge-inference}
|
|
\input{sections/09-provider-marketplace}
|
|
\input{sections/10-comparison-to-existing-platforms}
|
|
\input{sections/99-conclusion}
|
|
|
|
\end{document}
|