Built a full-stack probabilistic forecasting platform for the 2026 FIFA World Cup that estimates match outcomes and runs 20k–50k Monte Carlo simulations of the 48-team tournament to produce win/draw/loss, group-winner, advancement, and champion probabilities.

Combined an Elo logistic baseline, a Dixon-Coles Poisson score model, and a gradient-boosted match model into a weighted ensemble, with market-prior shrinkage to detect pricing edges.

Compared model probabilities against live Kalshi and Polymarket prices on a market-edge board and tracked a paper-trading ledger (CLV, ROI, hit rate) with liquidity/spread filters and fractional-Kelly sizing — validated with chronological backtests at ~60% top-pick accuracy and log-loss well better than random. Forecasts refresh and redeploy automatically via GitHub Actions.

Built a full-stack web app to streamline Princeton Neuroscience Institute course waitlists and enrollment, developed with a team of 5.

Implemented role-based access for different user types and queue logic to manage waitlist position and the enrollment flow.

Designed a responsive React frontend with Tailwind CSS, backed by a Flask API and a PostgreSQL database, deployed on Render.

Built an AI-powered financial agent that simulates equity investment decisions using LLM reasoning and function calling.

Integrated live market data (yFinance) with OpenAI/LangChain reasoning and SQL storage for analysis, memory, and trade rationales.

Exposed the agent through a Streamlit interface for interactive investment simulations.

Built a Flutter/Firebase productivity app that pairs full task management with a casino-style motivation system.

Implemented authenticated, Firestore-backed todos with due dates, recurring schedules, subtasks, categories, notes, reminders, and file attachments (Firebase Storage), plus list and calendar views with search, sort, and filter.

Designed a gamified rewards loop where completing tasks earns roulette spins and “House Chips” for table bets or deadline-tied “Boss Bets,” backed by transactional chip balances and confetti animations.

Built an AI investment agent that fetches real-time stock data (via yfinance) and uses a conversational LLM (Hugging Face) to answer natural language questions about any stock ticker.

Implemented a Gradio web interface so users can chat (like ChatGPT) and dynamically change the ticker (e.g., "/ticker TSLA") to get instant insights.

Integrated real-time market data (Open, High, Low, Close, Volume) and conversational AI to generate personalized stock summaries and investment insights.

Developed a full-stack web application using Python, Flask, and the seam carving algorithm, enabling users to upload images, specify resizing parameters, and download optimized results.

Leveraged Pillow and NumPy to implement intelligent content-aware resizing by dynamically removing horizontal and vertical seams, preserving critical visual details.

Analyzed Chicago ride-sharing data in RStudio, using the sf package for spatial analysis and infer package for hypothesis testing.

Applied LASSO and Ridge regression in tidymodels for predictive modeling and cluster analysis for pattern detection.

Created a Java program to analyze video data of polystyrene beads undergoing Brownian motion, applying image processing techniques.

Calculated Avogadro's number and Boltzmann's constant using Einstein's theory of Brownian motion.