Selected Work

Projects

A mix of quantitative research infrastructure and machine learning work. The through-line is validation: building tools that are honest about whether a result is real.

Over two years of independent research, I built a tick-level backtesting stack and used it to test 13 systematic trading strategies. Twelve were rejected. The one signal that reproduced was non-directional, and I reframed it as a risk-management input rather than alpha. The framework, not the strategies, is the real product.

Trendline & Options Automation

Python · mplfinance · options analytics

Scans a watchlist daily, fits support/resistance trendline channels to price history, and flags symbols that meet setup criteria. For each hit it computes a support "pillow," projected trendline/strike intersections, recent vs long-term volatility, and Fibonacci levels.

  • Automated channel fitting across many symbols per session
  • Projects where option strikes intersect fitted trendlines over time
  • Output feeds the Daily Charts page
Pythonpandasmplfinance

Tick-Level Backtesting & Validation Framework

Python · polars · NumPy · SciPy

A research engine for evaluating systematic strategies with execution realism and honest statistics. Built to make it hard to fool myself.

  • Tick-exact fill and stop/target resolution, avoiding the optimistic bias of bar-based tests
  • Null controls: re-run a strategy with its signal removed, so a result must beat its own placebo
  • Out-of-sample, cross-instrument, and cross-period validation with unit-invariant metrics
  • Quantified and corrected a bar-resolution simulation bias that inflates stop-based backtests
polarsNumPySciPystatistics

Market-Data Pipeline & Options Analytics

Python · Databento · CME feeds

Ingests raw CME exchange trade data, stitches front-month contracts across rolls, and computes options analytics on top.

  • Memory-bounded streaming over 10+ GB of tick data on commodity hardware
  • Black-76 pricing and implied-volatility inversion
  • Dealer-gamma exposure modeling as a volatility regime signal
  • Empirically verified exchange aggressor-side conventions against the raw data
Databentooptionsmicrostructure

Production Computer Vision

PyTorch · TensorFlow · OpenCV

A decade of applied CV, most recently leading a team shipping detection, classification, tracking, and segmentation into a production security product running on live video.

  • Real-time inference optimization for throughput and reliability in the field
  • CNNs for plant disease detection; Siamese networks for verification
  • Behavior classification and motion prediction (postdoctoral research, U.S. Army Research Laboratory)
PyTorchTensorFlowdetectiontracking