Projects
Selected public repositories grouped by the type of problem they solve. This page is curated for quantitative research and trading roles: the emphasis is on market microstructure, pricing, forecasting, and production-minded engineering. Repositories with missing or placeholder documentation are intentionally omitted to keep the portfolio focused and high-signal.
Quant Trading & Market Microstructure
Execution-aware systems built around live crypto market data, order book analytics, and simulated trading logic.
Real-Time Market Data & Market Making
Built a live market microstructure dashboard and simulated market-making engine to study quoting logic, queue position, fill quality, and short-horizon risk on crypto order books.
- Streams live order book and trade data, with spread statistics and configurable depth views
- Implements
mid,top,level,microprice, andqueue_awarequote placement modes - Tracks adverse selection through 1s/5s/10s markout, queue diagnostics, and fill analytics
- Adds risk controls on inventory, notional exposure, volatility gating, and max loss
Market Data Aggregation & Paper Trading API
Built a multi-exchange crypto data router and paper-trading API that normalizes live feeds from Binance and OKX into a unified execution and monitoring layer.
- Maintains persistent WebSocket connections to Binance Futures and OKX perpetual swaps
- Serves best touch, trades, klines, and EWMA streams through FastAPI and authenticated WebSocket endpoints
- Supports JWT auth, multi-asset balances, limit/market/IOC orders, and SQLite-backed persistence
- Detects cross-exchange negative spreads and surfaces simple arbitrage opportunities in the dashboard
Derivatives Pricing & Numerical Methods
Pricing engines focused on model validation, numerical stability, Greeks, and interactive diagnostics.
Basket Option Pricing Engine
Designed a multi-asset basket option pricer combining analytical moment-matching and Monte Carlo reference pricing with real market data integration.
- Implements Brigo-style moment matching plus a correlated Monte Carlo engine with control variates
- Supports both flat-parameter and term-structure frameworks, with ECB EURSTR and Bloomberg volatility inputs
- Achieves about 98.6% variance reduction in the reference Monte Carlo configuration
- Includes 15 automated tests covering pricing consistency, convergence, and economic properties
Monte Carlo Option Pricing Engine
Built an interactive options pricer centered on Monte Carlo simulation, Longstaff-Schwartz for American exercise, and research-style diagnostics.
- Prices European, binary, and American options with vectorized Monte Carlo and least-squares regression
- Computes Delta, Gamma, Vega, Theta, and Rho with Common Random Numbers and confidence intervals
- Benchmarks against Black-Scholes and a trinomial tree reference implementation
- Visualizes GBM paths, convergence rates, strike profiles, and payoff sensitivity in real time
Trinomial Tree Options Pricer
Implemented a trinomial tree pricer for European and American options with dividends, Greeks, and full tree visualization.
- Recombining tree engine with discrete dividend handling and Black-Scholes comparison
- Computes Greeks numerically and exposes convergence, precision, and execution-time studies
- Adds pruning logic to manage node explosion and improve runtime on deep trees
- Ships as a Streamlit app with Plotly-based node inspection and CSV export
Trinomial Option Pricer in C#
Translated and validated a Python trinomial pricer into C#, with a focus on numerical fidelity, speed, and cleaner object-oriented structure.
- Prices European and American calls/puts with discrete dividends and full Greeks
- Converges to within 1 cent of Black-Scholes at 400 steps
- Runs large trees of roughly 80,000 nodes in under 2 seconds
- Uses a modular architecture around market, option, tree, node, Black-Scholes, and Greeks classes
Econometrics, Forecasting & Applied ML
Mixed-frequency forecasting, financial machine learning, and time-series model comparison with an emphasis on methodology and validation.
GDP Forecasting with MIDAS Regressions
Replicated and extended ADL-MIDAS models to forecast quarterly US GDP using daily financial variables and mixed-frequency econometrics.
- Reproduces the Andreou-Ghysels-Kourtellos framework with Exponential Almon weights and recursive out-of-sample forecasts
- Uses Bloomberg data across equities, rates, commodities, FX, and macro indicators
- Adds a two-beta lag/lead extension to separate past information from nowcast information
- Compares specifications with RMSFE-based evaluation and forecast-combination logic
Financial Allocation Performance Prediction
Built a notebook-based machine learning workflow to predict the sign of short-horizon allocation returns on anonymized panel data.
- Uses leakage-safe chronological validation instead of random splits on time-dependent financial data
- Engineers features from 20-day return and signed-volume histories across 527k+ observations
- Compares Logistic Regression, dense nets, Random Forest, and LightGBM, with interpretability support
- Best rerun accuracy reaches 0.5246, while the final exported model favors LightGBM for a better generalization tradeoff
Quantitative Time Series Modeling
Exploratory time-series project focused on data quality, feature engineering, model comparison, and quantitative evaluation rather than a single black-box model.
- Studies trend, seasonality, anomalies, and residual structure on transactional sales data
- Compares naive baselines, ARIMA/SARIMA, and regression-style forecasting approaches
- Builds lag, rolling-statistic, and decomposition-based features
- Reports RMSE, MAE, and MAPE with attention to overfitting and robustness
Software Engineering & Production Systems
Supporting engineering projects that show API design, MLOps workflow thinking, deployment basics, and lower-level implementation discipline.
StackOverflow Tag Prediction - MLOps Pipeline
Turned a text classification problem into a small production-style ML platform with serving, orchestration, testing, and containerization.
- Generates BERT embeddings and trains a neural classifier for multi-label tag prediction
- Serves inference through a FastAPI endpoint with typed request/response validation
- Orchestrates training and inference jobs with Airflow and Docker Compose
- Includes modular code, automated tests, and artifact versioning for repeatable runs
C++ Matrix Library Project
Implemented a compact matrix library in C++ to reinforce low-level numerical programming, testing discipline, and linear algebra fundamentals.
- Supports matrix addition, multiplication, transpose, determinant, and matrix-vector products
- Uses row-major storage, dimension checks, and exception-safe APIs
- Compiles cleanly under strict flags with zero warnings
- Ships with basic and extended test suites plus small performance benchmarks
DevOps Fullstack Deployment Project
Built and deployed a small full-stack environment to practice containerized backend delivery and service orchestration.
- Containerizes PostgreSQL, a Java backend API, and an HTTP server acting as reverse proxy
- Manages services, networking, and startup flow with Docker Compose
- Covers persistence, environment variables, and container-to-container communication
- Demonstrates a foundation for shipping data or trading tools beyond local notebooks