Quantitative finance · ML · everything I build runs live

MSc Financial Modelling & Investment — University of Glasgow (Merit) · Adam Smith Skills Excellence Award
BCom Financial Analytics — Christ University (3.58 GPA)

I build quantitative tools end-to-end: from raw data ingestion through model validation to deployed interactive dashboards. My projects span derivatives pricing, recession forecasting, equity valuation, sports analytics, and regime detection — all with real data and live URLs.

📍 Glasgow, UK  ·  🔍 Open to market risk, quantitative analyst, and financial modelling roles  ·  📫 maanitkeyhem@gmail.com  ·  LinkedIn

→ Open app · Repo

Surface-specific Elo ratings built from scratch (clay, hard, grass, global) on 25 years of ATP and WTA data, feeding a Monte Carlo bracket simulator with 10,000 runs and bootstrap confidence intervals. Walk-forward backtesting gives 69% ATP / 67% WTA accuracy on clay. Simulate the full 128-player draw, run head-to-head matchups, or build your own custom bracket.

Python Elo Ratings Monte Carlo Logistic Regression Walk-forward Backtesting Streamlit

→ Open app · Repo

Real-time US Treasury yield curve decomposition (Nelson-Siegel: Level, Slope, Curvature) feeding a probit model outputting NBER recession probabilities at 3, 6, and 12-month horizons — pulling live FRED data on load. Validated with zero look-ahead bias.

Python Nelson-Siegel Probit Streamlit FRED API · AUC = 0.82 at 12m · Diebold-Mariano tested

→ Open app · Repo

Stochastic derivatives pricing engine: Black-Scholes analytics combined with 100,000-path Monte Carlo simulation (GBM, 95% CI bootstrapping) on live market data. Greeks rendered as interactive 3D surfaces across strike × time-to-expiry; IV smile and multi-leg payoff diagrams included.

Python Black-Scholes Monte Carlo (GBM) Dash yfinance · 100k simulation paths

→ Open app · Repo

Automated equity valuation platform pulling live financials via FMP API. Computes WACC with CAPM using a live FRED risk-free rate, runs a 5-year DCF with Gordon Growth terminal value, and outputs an equity bridge to intrinsic price per share. Includes 2D sensitivity heatmap (WACC × growth), Bear/Base/Bull scenario analysis, and comparable company multiples (EV/EBITDA, EV/Revenue, P/E, P/FCF) with one-click PDF and Excel export.

Python FMP API FRED API CAPM DCF Dash · Real-time price streaming during NYSE hours

→ Repo

End-to-end ML pipeline integrating 6 FRED macro series with 25 years of BTS Form 41 financial data across 8 US carriers. Engineered 44 features (macro lags, rolling trend signals, market share, Herfindahl index, seasonality dummies). SHAP explainability surfaces dominant drivers — trailing momentum, jet fuel prices, Q3 seasonality — as stakeholder-ready charts.

Python XGBoost SHAP Dash FRED API · R² = 0.77 OOS · 5-fold TimeSeriesSplit CV

→ Repo

Hidden Markov Model trained on 25 years of macro data (yield curve, VIX, CPI) detecting Expansion, Baseline, and Stress regimes. Correctly classified 2008 GFC and COVID-19 as stress periods; walk-forward backtesting confirms 2.6pp lower portfolio volatility vs S&P 500 benchmark.

Python Hidden Markov Models FRED API · Walk-forward backtesting · zero look-ahead bias

→ Repo

Production-grade SQL data warehouse ingesting EPA GHGRP facility emissions data (US Scope 1) and EU ETS verified installation records. Analytical layer covers sector benchmarking, intensity trends, emissions trajectory regression (LAG, REGR_SLOPE, PERCENTILE_CONT window functions), and percentile ranking — ~10k+ real installations.

Python PostgreSQL SQL EPA GHGRP EU ETS · Real data · MRV-aligned

→ Repo

Staggered difference-in-differences analysis of EU accession's macroeconomic effects across 23 countries using World Bank panel data. Found accession raises exports by +7pp, imports by +10pp, and capital formation by +4.5pp of GDP.

Stata DiD World Bank Panel Data · Published econometric methodology

Everything here is deployed with real data. If something is broken, open an issue — I check regularly.