projects
Four rigorous academic and applied research projects in quantitative finance, statistical modeling, and institutional AI strategy.
post-trade performance attribution framework
A production-grade Python framework designed to decompose and quantify execution costs across a corpus of 5,000+ simulated equity trades. The system implements the Implementation Shortfall (IS) model — the industry-standard methodology for measuring the gap between a theoretical decision price and actual realized execution cost.
The framework computes three primary metrics: arrival cost (the shortfall from decision-point mid to execution price), VWAP slippage (execution performance relative to the volume-weighted average price over the trading interval), and Z-score normalization to surface statistically significant execution outliers. Outputs are structured to support downstream analysis of broker performance and market impact models.
Built with an emphasis on auditability — every computation is reproducible from raw trade log inputs, with clear separation between data ingestion, metric calculation, and reporting layers.
machine learning for business — classification capstone
A rigorous model-comparison study benchmarking five classification algorithms — logistic regression, random forest, XGBoost, multi-layer perceptron, and a baseline naive Bayes — on a dataset of 50,000 financial transactions. The objective: identify which model architecture generalizes best under realistic conditions, not idealized benchmark conditions.
The defining contribution of this project was the design and execution of a forensic no-leakage experiment. During initial evaluation, performance metrics appeared implausibly high. Systematic investigation of the feature pipeline revealed that target-correlated features had been inadvertently included during preprocessing — a subtle form of synthetic data leakage that invalidated prior results across all five models. The experiment was redesigned from scratch with strict temporal and informational barriers, and the findings were reframed to foreground the detection methodology as a practitioner's guide to trustworthy ML evaluation in finance.
fintech capstone — ai adoption at jpmorgan chase
Led a cross-functional research team commissioned to map and analyze the AI/ML deployment landscape at JPMorgan Chase — one of the most sophisticated institutional adopters of machine learning in global finance. The scope: 300+ active use cases spanning trading, risk, compliance, consumer banking, and operations.
The team developed a taxonomy for categorizing AI deployments by maturity, domain, and business criticality — then synthesized findings into a dual-audience deliverable. Technical stakeholders received granular analysis of model architecture choices, data pipeline dependencies, and infrastructure constraints. Non-technical stakeholders received a strategic narrative about where AI creates durable value versus where its current limitations demand human oversight. Presented findings live to both audiences, fielding technical questions and strategic objections in real time.
pre-trade strategy optimization
Implemented and evaluated optimization algorithms for multi-scenario trading strategy problems under realistic operational constraints. The core challenge: given a target position and market liquidity profile, determine the execution schedule that minimizes market impact and timing risk while respecting participation-of-volume (POV) limits and fixed time horizons.
The project modeled multiple constraint regimes — varying POV caps, execution windows, and urgency parameters — and compared the resulting cost frontiers across scenarios. Outputs demonstrate the sensitivity of total execution cost to constraint tightness, providing a framework for pre-trade decision support that can be adapted to live order management workflows.
interested in the methodology behind any of these?
Get in Touch →