MC

Projects

Forecasting and software systems

Each project includes problem framing, technical approach, measurable outcomes, and implementation lessons.

HYDRA

Hybrid Deep-learning for Residual Analysis

Problem

National water models often struggle with local precision. During events like Hurricane Helene, NWM forecasts significantly underestimated peak flows in complex terrain, leaving communities with uncertain flood warnings.

Why it matters

Achieved up to 48% improvement in predicting streamflow across unregulated Appalachian watersheds. Live at hydramodel.ai. Manuscript in preparation for Water Resources Research (AGU).

Approach

  • Developed an advanced 1-million parameter hybrid GRU-Transformer model to catch residuals that physical models miss.
  • Integrated multi-source data (NWM v3.0+, ERA5, USGS) for a comprehensive view.
  • Designed leakage-safe evaluation to ensure performance holds up in real-time constraints.
  • Preparing manuscript for submission to Water Resources Research (AGU).

Stack

PyTorchTransformerHydraMLflowxarrayAWS Batch

Results

  • Up to 48% improvement in predictions across unregulated Appalachian watersheds.
  • Consistent improvement across varying forecast horizons and extreme storm events.
  • Demonstrated stability under seasonal distribution shifts.

What I learned

  • Hybrid architectures can correct physical model biases without overfitting.
  • Rigorous evaluation design is the difference between a demo and a product.
  • Balancing research goals with operational constraints mirrors how ML systems are built in production teams.

USAA Risk Services

Secure Production APIs for Financial Services

Problem

Internal teams needed a reliable, compliant way to access core risk data without navigating legacy complexity.

Why it matters

Built production APIs that partner teams across the organization now use daily. Worked closely with product owners, backend engineers, and internal stakeholders to make risk data accessible and self-serve.

Approach

  • Designed and deployed strict GraphQL APIs using Java and Spring Boot.
  • Enhanced internal observability tools by unifying disparate data sources.
  • Built responsive React dashboards to give stakeholders visibility into risk metrics.

Stack

JavaSpring BootGraphQLReactTypeScriptPostgreSQL

Results

  • Consolidated fragmented data access into a single self-serve API layer, eliminating recurring partner-team support loops.
  • Introduced schema-validated deployment gates that caught contract regressions before production.

What I learned

  • Clear interfaces (API contracts) allow teams to move fast safely.
  • Effective production systems require close collaboration between engineers, product owners, and end users.
Active Development

Harmony

Algorithmic trading systems for disciplined decision-making

Problem

Financial markets are noisy, adversarial environments where predictive accuracy alone is insufficient. Sustainable trading systems require disciplined evaluation, strict risk controls, and infrastructure that behaves predictably under stress.

Why it matters

A modular trading system with independent forecasting, risk, and execution components. Each module can be evaluated, replaced, or improved independently.

Approach

  • Design a forecasting and execution pipeline that survives regime shifts.
  • Emphasize risk management over raw returns.
  • Build components that can be evaluated independently and replaced safely.
  • Treat trading as a systems engineering problem, not a single model.

Stack

PythonPyTorchpandasNumPyPostgreSQLDocker

Current Scope

  • Market data ingestion and normalization pipeline.
  • Forecasting models for short- and medium-horizon signals.
  • Risk engine enforcing position sizing and drawdown limits.
  • Backtesting framework with leakage-aware evaluation.
  • Execution simulation via paper trading.

Design Principles

  • No model is trusted without adversarial backtesting.
  • Risk constraints override model confidence.
  • Components must fail safely.
  • Performance claims require statistically sound evaluation.