Graduate student in Data Science (Quantitative Finance) at Columbia University with a background in Distributed Systems (IP Paris) and Engineering (Telecom SudParis).
I specialize in Quantitative Research, Systematic Asset Allocation, Machine Learning for Finance, and AI-driven Financial Engineering.
- QRT x École Normale Supérieure Datathon: Engineered next-day return direction prediction pipeline.
- Purged Cross-Validation: Demonstrated Purged-CV + XGBoost outperforming benchmark architectures on held-out leaderboard data.
- Matrix Overlap Reconstruction: Recovered chronological order of in-sample shuffled dates from 20-day return windows via nearest-neighbor overlap matching.
- Equity Directional Prediction: Built a stacked LSTM architecture targeting daily directional movement in US Biotech equities.
- Dollar-Bar Sampling: Implemented de Prado dollar-bar event sampling to stabilize predictions during volatility shocks.
- Feature Stability & Hardening: Applied distributional moment profiling, winsorization, and z-score recalibration.
- 13F Theme Timing & Early-Mover Detection: Built a pipeline analyzing 3,000+ hedge fund managers to detect early-mover skill using SEC DERA 13F filings (2013–2026) and 5,200-firm Preqin-to-CRSP mappings.
- Hierarchical Modeling: Estimated theme conviction via hierarchical random slopes; identified top-quintile managers across ~800 pairs after FDR control.
- Four-Factor Style Model & Risk Optimization: Developed a multi-asset style model (Carry, Value, Momentum, Quality) across US ETF asset classes (equities, FICC, commodities).
- Orthogonalization & Signal Isolation: Created cross-sectional Z-scores with sector demeaning, isolated pure style premia via OLS market-premium orthogonalization.
- Mean-CVaR LP Formulation: Formulated and solved Rockafellar-Uryasev Mean-CVaR Linear Programs (CVXPY/SCS) across a 60-month rolling walk-forward backtest.
- Macroeconomic LLM Platform: Designed a domain-engineered platform extracting discretionary timing and sizing signals from unstructured macroeconomic publications.
- Noise Reduction & Anomaly Engine: Constructed a z-score gating engine using log-share deviations (MAD) on macro indicators, reducing signal noise and false positives.
