papers/zoo-educational-ai/sections/06-system-architecture.tex

\section{System Architecture}

\label{sec:architecture}

\subsection{Decentralized Infrastructure}

ZooEd is deployed on the Zoo Network with a hybrid architecture optimized for the connectivity constraints of conservation field sites:

\begin{figure}[H]
\centering
\begin{tikzpicture}[
    node distance=1.5cm,
    box/.style={rectangle, draw, rounded corners, minimum width=3cm, minimum height=0.8cm, text centered, font=\small},
    edge/.style={->, thick}
]
    \node[box, fill=blue!10] (client) {Mobile Client};
    \node[box, fill=green!10, right=2.5cm of client] (edge) {Edge Node};
    \node[box, fill=orange!10, right=2.5cm of edge] (network) {Zoo Network};
    \node[box, fill=red!10, below=1cm of client] (offline) {Offline Cache};
    \node[box, fill=purple!10, below=1cm of edge] (sensor) {Sensor Gateway};
    \node[box, fill=yellow!10, below=1cm of network] (ipfs) {IPFS Content};

    \draw[edge] (client) -- (edge);
    \draw[edge] (edge) -- (network);
    \draw[edge] (client) -- (offline);
    \draw[edge] (sensor) -- (edge);
    \draw[edge] (network) -- (ipfs);
    \draw[edge, dashed] (offline) -- (edge);
\end{tikzpicture}
\caption{ZooEd system architecture showing the three-tier deployment model.}
\label{fig:architecture}
\end{figure}

\begin{itemize}[leftmargin=*]
    \item \textbf{Mobile Client:} Progressive web application with offline-first design, capable of running the learner model and basic assessments without connectivity.
    \item \textbf{Edge Node:} Solar-powered computing units deployed at conservation sites, running the adaptive engine and caching content. Each edge node serves up to 200 concurrent learners.
    \item \textbf{Zoo Network:} Provides decentralized storage for content (IPFS), on-chain consent management, and coordination between edge nodes.
\end{itemize}

\subsection{Data Governance}

Learner data governance follows the Zoo Network's consent framework:

\begin{enumerate}[leftmargin=*]
    \item \textbf{On-Chain Consent:} Each learner registers a consent profile specifying data sharing preferences, stored as an immutable on-chain record.
    \item \textbf{Differential Privacy:} All aggregate analytics apply $(\epsilon, \delta)$-differential privacy with $\epsilon = 1.0$, $\delta = 10^{-5}$.
    \item \textbf{Data Minimization:} The learner model operates on feature vectors derived from interaction patterns; raw content of free-response answers is discarded after assessment scoring.
    \item \textbf{Right to Deletion:} Learners can trigger cryptographic erasure of their interaction history while preserving aggregate model parameters.
\end{enumerate}

\subsection{Incentive Mechanisms}

ZooEd integrates with the Zoo token economy to incentivize ecosystem participation:

\begin{table}[H]
\centering
\caption{Token incentive structure for ZooEd participants.}
\label{tab:incentives}
\begin{tabularx}{\textwidth}{@{}llX@{}}
\toprule
\textbf{Role} & \textbf{Reward} & \textbf{Mechanism} \\
\midrule
Content Author & 50--500 ZOO & Per accepted learning module, scaled by language scarcity and domain demand \\
Translator & 10--100 ZOO & Per verified translation, with quality multiplier from back-translation score \\
Peer Instructor & 5--50 ZOO & Per tutoring session, scaled by learner improvement \\
Learner & 1--10 ZOO & Per mastered objective, claimable after demonstrated retention at 30 days \\
\bottomrule
\end{tabularx}
\end{table}

% ============================================================
% 7. Experimental Evaluation
% ============================================================