Selected ML Systems
Financial ML Pipeline
Problem: Financial models require rigorous temporal validation to avoid lookahead bias and account for regime shifts in non-stationary markets.
Approach: Built end-to-end pipeline with walk-forward splits, regime-aware cohort analysis, and comprehensive experiment tracking. Implemented versioning for data, code, and model artifacts to ensure reproducibility.
Outcome: Production pipeline enabling valid out-of-time performance assessment for trading strategies. Framework prevents common pitfalls in financial ML including data leakage and overfitting to historical regimes.
Python
XGBoost
LightGBM
Docker
Statistical Validation Framework
Problem: Standard cross-validation underestimates variance in financial predictions with autocorrelated outcomes and non-IID data.
Approach: Developed bootstrap and permutation testing methods adapted for time-series. Implemented confidence interval estimation and robustness metrics under different market regimes.
Outcome: Framework enabling statistically sound model selection and honest performance reporting. Prevents overconfidence in backtest results and supports regulatory review.
Statistical Inference
Time-Series
Bootstrap Methods
Feature Engineering Service
Problem: Feature drift between training, backtesting, and live deployment creates silent model degradation that compounds over time.
Approach: Created configuration-driven feature computation layer with versioning, point-in-time correctness guarantees, and support for both historical replay and streaming data.
Outcome: Eliminates lookahead bias and ensures reproducibility across model lifecycle. Single source of truth for features reduces debugging time and prevents production incidents.
Feature Engineering
Data Pipelines
Production ML
Background
Independent ML Engineer / Researcher (2020–Present): Production ML systems for algorithmic trading and quantitative finance. Built reinforcement learning pipeline processing 7,876 ticks/sec; conducted research on market predictability with honest reporting of null results.
Founder & CTO, Peoples' FinTech (2017–2020): Led development of quantitative trading platform integrating ML models with event-driven architecture.
Earlier roles: Programming Manager at Concord Wealth Management (2007–2008); Engineering and modeling work at Daren Labs Scientific and Anitani Solutions (2003–2017).
Resume (PDF)