Jack Schiavo

Physics & Quantitative Economics — Saint Joseph's University

I build software that runs real businesses
and models real risk.

Six projects from this summer: a live restaurant-analytics SaaS with paying customers, a sister product for wedding & event venues, a quantitative portfolio optimizer, a battery storage arbitrage calculator on real energy-market data, and two quantum-computing studies — one benchmarked against classical machine learning, one a live options-pricing calculator measuring a real quantum speedup.

See the projects Get in touch

Projects

Flagship — live, paying customers

RestaurQuant

A restaurant analytics platform that turns raw POS data (Toast, Square) into the numbers owners actually need to run the business: prime cost, menu-item margin quadrants, labor cost against sales, and revenue trend analysis — with an AI assistant that can answer plain-English questions about the data. Solo-built and solo-operated end to end: product, pricing strategy, live Stripe billing, App Store submission, SEO, and outreach to real restaurants.

  • Live web app + native iOS app on the App Store
  • Real subscription billing (Stripe), self-serve signup, multi-location support
  • Two-tier pricing differentiated by POS integration cost (Toast vs. Square)
  • Direct POS integrations, nightly data sync, role-based access control
restaurquant.com ↗

Live — sister product

Gala by RQ

A per-event profitability tool for wedding venues and banquet halls. Most venues can tell you what an event billed — Gala tells them what it actually made, netting booking revenue against staffing, catering, and vendor costs, with an optional QuickBooks Online connection for venues that already do job costing. Built as a standalone product with its own landing page, billing, and design identity.

  • Full self-serve product: signup, dashboard, CSV/Excel bulk import, 14-day trial
  • QuickBooks Online integration for automatic per-event financials
  • Hybrid pricing: flat fee for manual entry, usage-based when connected to real books
restaurquant.com/gala ↗

Live — quant & quantum computing

QUBO Portfolio Optimizer

An interactive portfolio construction tool that formulates stock selection as a cardinality-constrained mean-variance QUBO (Quadratic Unconstrained Binary Optimization) over the real S&P 500 — the same problem structure used in quantum-annealing and QAOA research, solved here at production scale. Given a risk tolerance, capital, and number of holdings, it returns a concrete portfolio, a Monte Carlo range of outcomes, descriptive risk flags (concentration, correlation, volatility), and a hedge-sizing calculator — deliberately built as decision-support analytics, never a buy/sell signal.

  • Real S&P 500 universe (496 stocks), classical simulated-annealing solver
  • Existing-holdings-aware: new picks avoid correlating with what a user already owns
  • Automated weekly market-data refresh and a running paper-trading simulation
Live app ↗

Live — energy markets & optimization

Battery Storage Arbitrage Calculator

An interactive tool that answers a real energy-trading question: what could a grid-scale battery actually earn trading real day-ahead electricity prices? Deliberately built to need zero forecasting — NYISO publishes all 24 hours of next-day prices at once, so optimal battery dispatch against them is a perfect-information linear program, not a prediction problem. Every input change re-solves the real optimization live, against a full year of real NYISO market data.

  • Live LP solve per request (SciPy/HiGHS, ~0.4s) — not a precomputed lookup table
  • Results land in the range of published real-world battery storage revenue ($20-90+/kW-year)
  • Tunable power, duration, and round-trip efficiency; charge/discharge dispatch charted live
Live app ↗

Live — quantum computing & derivatives pricing

Option Pricing via Quantum Amplitude Estimation

An interactive calculator that prices real European call and put options using Quantum Amplitude Estimation and measures, live, whether it actually needs less computational work than the classical Monte Carlo method finance normally uses. Every price is computed on request by a real Qiskit quantum-circuit simulation, cross-checked against the closed-form Black-Scholes price, alongside a reference chart proving the quantum method's cost scales close to its theoretical quadratic advantage over classical sampling.

  • Measured result: quantum cost scales as ε-1.04 vs. classical ε-2.00 — a real, not assumed, speedup
  • Also validated on real IBM quantum hardware, within a fraction of a cent of the simulator
  • Live call/put pricer, adjustable strike/volatility/expiry/precision, real quantum circuit evaluations per request
Live app ↗

Live — derivatives & risk management

Derivatives Risk Dashboard

An options risk tool built on real listed contracts, not textbook parameters: live Yahoo Finance option chains, Black-Scholes Greeks solved from the current bid/ask (Yahoo's own implied-vol field is sometimes stale, so this cross-checks it), an implied-vol surface, a multi-leg position builder with a scenario payoff diagram, and a historical delta-hedge backtest that only ever looks backward at real price paths — the same practical, everyday counterpart to the quantum option-pricing work below.

  • Every chain, Greek, and payoff diagram computed live per request against real quotes
  • Implied vol solved from bid/ask mid with a floor check that catches stale quotes Yahoo's own field misses
  • Delta-hedge backtest never uses a future price — same no-forecasting discipline as the battery arbitrage tool
Live app ↗

Research study

Quantum Kernel Electricity-Price Forecasting

A rigorous benchmark of quantum kernel methods against classical machine learning on real electricity market data — twelve months of NYISO day-ahead price and load data for the NYC zone. Two PennyLane quantum kernels (angle-embedding and an IQP-style feature map) were tuned through a twelve-configuration hyperparameter sweep and compared against a tuned classical RBF-SVR, gradient boosting, and an LSTM, with bootstrap confidence intervals on every result rather than a single point estimate.

  • Result: quantum kernels are statistically indistinguishable from a tuned classical SVR
  • Found and corrected a GBM overfitting trap that inflated error by 2.5x before tuning
  • Built on real public NYISO market data, fully reproducible pipeline

Research study — quantum computing & energy markets

Forecasting, Hedging & Pricing Grid Congestion

Three studies, one real instrument: Financial Transmission Rights, the standard hedge against grid congestion. A forecasting model predicts a real NYISO congestion spread, a portfolio optimizer picks which of 110 real contracts to hold, and a quantum option-pricing model prices a right on one of those spreads — wired to the forecasting model's own live prediction, not a stand-in number.

  • Pricing result: quantum needed ~5,000x fewer computations than classical Monte Carlo at high precision
  • Forecasting result: quantum kernels tie a tuned classical model; a plain LSTM beats both
  • Found the 110-contract universe has only ~11 truly independent sources of risk — and swapping in a real forecast changed which contracts the optimizer recommended
Read the full study →

Research study — quantum computing & derivatives pricing

Swing Option Pricing via Quantum Amplitude Estimation

A swing option stacks several exercise dates on top of a vanilla option, all fighting over one shared volume limit — nobody had pointed Quantum Amplitude Estimation at that combination before (checked twice, not assumed). Three actual builds, not one: an exact-but-exponential circuit, an efficient version that looked like a win until the total cost said otherwise, and a fix that keeps the real insight while dropping the part that made things worse.

  • Found and fixed a real bug — a decision made before its own price was known — caught only because an independent classical method disagreed
  • A structurally cheaper circuit (verified: Markovian policy, evenly-spaced grid) turned out to need so many more amplitude-estimation queries that it was worse than the original — caught by finishing the total-cost math, not stopping at the flattering number
  • Fixed version is ~620,000× cheaper than the original at longer contracts, but still loses to classical Monte Carlo — the crossover is decided by precision scaling, not contract length
  • Real IBM hardware run on the smallest possible contract measured much closer to pure noise than to the correct price — confirms, on a real chip, that the circuit is too gate-deep for today's hardware even at its smallest size
Read the full study →

About

I'm a rising senior studying Physics and Quantitative Economics (minor in Math) at Saint Joseph's University, researching quantum computing with an interest in energy markets, power systems, and commodities trading. I also work part-time on a restaurant floor, which is where RestaurQuant started, from seeing the gap between what a POS system logs and what an owner actually needs to know.

What connects these projects is the same instinct applied to different domains: take a real decision-making problem — restaurant margins, event profitability, portfolio risk, price forecasting — and build something that gives a straight, honest answer instead of a dashboard full of noise.