Matthew Casertano
Co-founder of ARGUS, building AI-powered AR guidance that helps industrial workers complete complex tasks without errors. Previously at Bridgewater Associates and NextEra Energy. Studying CS and Economics at Caltech.
Things I've Built
ARGUS
Real-time AR guidance and verification for industrial workers
The Problem
Industrial errors cost billions annually. A technician misidentifies a critical component during turbine maintenance — the entire power plant goes offline. An assembly line worker misaligns a part in an aircraft fuselage — expensive rework, flight delays, or, in the worst case, catastrophe. Training is expensive, doesn't stick, and can't cover every edge case. Paper checklists and videos don't work when you need both hands.
Our Solution
ARGUS provides step-by-step AR guidance overlaid on the physical workspace, then verifies each step was completed correctly before letting workers proceed. We use a hybrid verification approach that combines Vision-Language Models for semantic reasoning with traditional computer vision for deterministic reliability.
The result: Workers get intelligent guidance that understands context, backed by verification that actually works in production environments.
Technical Innovation
The fundamental challenge in AR verification is that Vision-Language Models alone are unreliable for safety-critical tasks. We solved this by building a hybrid system that fuses VLMs with YOLO/OpenCV:
- VLMs handle semantic reasoning and edge cases that traditional CV misses
- Deterministic CV provides guaranteed reliability for standard operations
- Custom synthetic data pipeline generates thousands of photorealistic labeled training images
- Result: Error rates dropped from >40% to <3% in lab testing, making the system actually deployable
Early Traction
Built in partnership with Caltech's CAST Lab under Professor Morteza Gharib. Active customer discovery in manufacturing, aerospace, and energy.
Other Projects
🌡️ Climate Comfort Index
Ever wondered where in the US matches your ideal weather? This web app takes your climate preferences and maps them across 3,000+ counties.
How it works: Input your ideal temperature ranges, humidity preferences, sunshine needs, and love (or hate) for snow. The algorithm calculates a "comfort score" for each county and generates a choropleth map.
🧬 NMRSuite
During my time at UMD's Fushman Lab, I built a web gateway for analyzing protein dynamics through NMR data. I took complex molecular modeling algorithms and made them accessible to researchers worldwide.
Impact: Published in Gateways 2021 and presented at the NSF-funded SGCI Conference. Now used by structural biology researchers to understand how proteins move and function.
Technical Challenge: Translating cutting-edge research algorithms into a user-friendly web interface that handles complex scientific data processing.
📈 Risk Parity Portfolio Analysis
Most portfolios are dominated by stock risk. What if we allocated based on risk contribution instead of dollars? I tested this approach across 18 years of market data, including three major crises.
Key Finding: A Stock-Bond-Gold risk parity portfolio achieved a 0.98 Sharpe ratio while limiting maximum drawdown to just 20.2% - dramatically outperforming traditional allocations during market stress.
Inspired by Bridgewater: Building on Ray Dalio's risk parity framework, I wanted to test how this approach holds up recently. It proved particularly effective during both the 2020 COVID crash and 2025 tariff crisis, especially when gold was added as a third asset.
🇨🇳 China's AI Strategy
While most attention was on the chip race, we identified how China could play a different game: leveraging open-source AI and legacy chip dominance instead of matching US semiconductor capabilities.
Core Insight: Rather than chasing TSMC's 3nm chips, China is positioning to control 40% of legacy chip production by 2027.
Thesis Validated: One month after our research, DeepSeek shocked the market in a way that matched our predictions: breakthrough AI performance with legacy H800 chips and open-source strategies.
🏆 Telecom Churn Prediction Competition
In Caltech's 2024 CS 155 Machine Learning competition, we were asked to predict telecom customer churn with noisy, real-world data.
My approach: I focused on feature engineering, creating domain-specific interaction terms that I thought could better capture patterns in customer behavior.
Result: Won 1st out of 58 teams.
🧠 Neural Autopilot Parameter Analysis
Parameter space analysis for a novel behavioral model explaining power laws in human behavior. Validated the potential for ground truth recovery of these models as part of research in the Colin Camerer lab at Caltech.
Research Impact: Contributed to understanding how habit formation and goal-directed behavior create bursty patterns in human activities like gym attendance and social media usage.
Technical Achievement: Conducted stability analysis across 1000+ parameter combinations and developed visualizations to understand model behavior.
Experience
Researched FX markets and formed systematic hypotheses on currency dynamics. Tested hypotheses with historical data. Immersed in formal training covering macroeconomics, analytical frameworks, and systematic modeling.
Conducted due diligence on AI/ML and enterprise SaaS investments. Built automated sourcing systems and led projects on edge computing and AI for customer service, presenting findings to executive leadership.
Developed NMRSuite for protein dynamics analysis. Published research and presented at NSF-funded conferences. Bridged the gap between cutting-edge science and practical tools.
Built nonprofit from scratch during COVID-19, connecting teen volunteers with vulnerable seniors needing grocery deliveries and companionship. Scaled to 31 chapters across the US and Canada with hundreds of volunteers making nearly 4,000 deliveries over three years.
Media Coverage: Featured in CNN, Good Morning America, NPR, The Washington Post, People Magazine, and PBS. Demonstrated ability to build and scale operations from zero.
Education
GPA: 4.1/4.0 • Research under Professor Morteza Gharib on AI-powered AR verification systems and Professor Colin Camerer on behavioral economics and LLM behavior in experimental markets • Selected for Housing Working Group to redesign campus housing lottery • 2025 Outstanding Leadership Award • Active in Quantitative Finance Club, CS Club, Poker Club, Sunday League Soccer
GPA: 4.82 • SAT: 1600 • Captain of Math Team & Financial Literacy Club • Lecturer for Computer Team & Debate Team