Joshua Caiata

Hey, I'm Josh!

I'm a master's student in Computer Science at the University of Waterloo, where I research multi-agent systems and decision-making. My work draws on ideas from artificial intelligence, economics, and machine learning, and focuses on empirically evaluating and designing decision-making systems.

Publications

What Voting Rules Actually Do: A Data-Driven Analysis of Multi-Winner Voting

Joshua Caiata, Ben Armstrong, Kate Larson

We introduce a data-driven framework for benchmarking decision rules using empirical performance rather than worst-case guarantees, and show that learned policies can outperform traditional rule-based systems in practice. This paper was accepted to AAAI 2026.

Link to Paper

Procedural Fairness in Multi-Agent Bandits

Joshua Caiata, Carter Blair, Kate Larson

We analyze how influence is distributed in multi-agent systems, using bandits as a test case. We show that standard optimization policies centralize control and dilute agent influence. We introduce an equal-influence framework that preserves performance across objectives with minimal efficiency loss, reframing governance as a core system design choice. This paper was accepted to the International Conference for Autonomous Agents and Multi-Agent Systems (AAMAS) 2026 as an extended abstract.

Link to Paper

Projects

CalTrack

SWIFT, SWIFTUI, SQL

A calorie tracker for iOS! Counts your burned calories using HealthKit to sync with Apple Health, and uses the FoodData Central API to find nutritional information for various foods.

OkCupid Predictor

PYTHON, PYTORCH, SCIKIT-LEARN, XGBOOST, MATPLOTLIB

Trained various models to predict relationship statuses using an OkCupid dataset. Trained a decision tree, support vector machine (SVM), neural network, and XGBoost.

Mazur

JAVASCRIPT, REACT, HTML, CSS

Built a random maze generator that can also solve the maze using Dijkstra's, A*, breadth-first, and depth-first search algorithms. This uses the Hunt and Kill algorithm to generate the mazes.

NYC Ride Time Predictor

PYTHON, XGBOOST, SCIKIT-LEARN, MATPLOTLIB, PANDAS

I did this project to learn about Data Science. I used Matplotlib to plot characteristics of the data, and used XGBoost to predict the ride times.

Revolt Fashion

JAVASCRIPT, REACT, HTML, CSS, COMMERCE.JS, STRIPE

Built a small, high-end (demo) e-commerce website to learn how to use React and API’s. Note: As of July 2023, commerce.js is not rendering images on Safari.

Code History

SWIFT, SWIFTUI

Something I built while I was learning Swift! It has a couple questions about computer science history, and it will calculate your score when you complete the quiz.

Personal Portfolio

JAVASCRIPT, REACT, HTML, CSS

This is my portfolio, the one you are viewing right now. I did all of my own custom CSS, including the CSS for the animations. The site is responsive, so it works for any screen sizes.

Park Passport

JAVASCRIPT, REACT, HTML, CSS

A ride tracker app for all of your favourite Disney theme parks! Count how many times you have been on each ride in each park. This works for all parks in the continental United States.

Basecamp Finance

PRODUCT MANAGER

This was a project I helped on where I was a Product Manager. I conducted market research, developed the product roadmap, wrote product requirements, and designed the wireframes.

Straights - Command line game

C++

This is a fun little command line card game built using object-oriented programming principles such as encapsulation, abstraction, polymorphism, and inheritance.

Code Compiler

RACKET (LIKE LISP)

This is a compiler for a subset of the C language that performs scanning, parsing, context-sensitive analysis, code generation, linking, and assembling.

Hangman

PYTHON

This is a small hangman game I built when I was learning how to code. It picks a word, takes user input, and will continue until you win or lose the game!