Aman Kumar Jethani

Enterprise Data Architecture & Expected Return (ER) Modeling Orchestrated a resilient data pipeline migration from S&P 500 to LSEG within Snowflake, ensuring continuity for quantitative modeling; optimized SQL joins to evaluate 5K equities in few seconds and established automated cloud cost-reduction alerts. Engineered a production-grade Point-in-Time (PiT) data system leveraging time-travel concepts to accurately map corporate actions and historical restatements, completely eradicating look-ahead bias in ML models. Researched and engineered ∼150 custom fundamental alpha signals; achieved ∼23% RMSE reduction and outperformed baseline in ∼70% of production scenarios through rigorous feature selection and cross-validation. Formulated a custom CatBoost loss function tailored to directional return prediction, delivering ∼27% improvement in directional accuracy over standard objectives. Integrated short- and long-term ER models into a monthly ETF portfolio, executing fundamental strategies that surpassed the S&P 500 (SPY) benchmark by 0.5% monthly and captured a maximum 4% annual excess return. Quantitative Portfolio Optimization & NLP Strategies Spearheaded a thematic sector-rotation strategy on the S&P 500 by applying NLP pipelines on SEC filings and earnings call transcripts; dynamically selected ∼200 equities per cycle, generating ∼5% annualized alpha in rigorous walk-forward backtesting. Combined factor-based portfolio construction with NLP-derived signals to improve signal diversity and reduce factor crowding in long-short equity portfolios. Auto-Scaling MLOps & Infrastructure Architected end-to-end auto-scaling pipelines spanning data ingestion, feature computation, model training, and inference serving; reduced training cycle times by 30–40% through parallelization and resource-aware scheduling. Implemented automated drift-detection and retraining triggers, enabling continuous model adaptation and sustaining a 2–3% performance lift over static baselines in production. Advanced Analytics & Sentiment Analysis Built a real-time financial news scraping and classification bot powered by OpenAI API, autonomously detecting and quantifying litigation risks, M&A activity, and adverse market sentiment to generate actionable event-driven signals.

BNPL Fraud & Risk Engineered a BNPL fraud detection model extracting ∼300 custom features, achieving a 65% GINI coefficient — a 10% lift over the legacy LightGBM baseline. Optimized BNPL risk decisioning algorithms, slashing early-stage delinquency (DOB22) by 38% to a record low of 6.17% and driving substantial annualized cost savings.