RAG-based Q&A system that answers board game rules questions with citations, multi-hop reasoning, and honest uncertainty handling. Hybrid search combining Pinecone dense embeddings and BM25 sparse retrieval with Reciprocal Rank Fusion. Three-tier response system: direct answers with citations, multi-hop chain-of-retrieval for complex queries, and honest uncertainty when confidence is low. Includes query rewriting via Claude Haiku, cross-encoder reranking, and citation verification. Deployed on Railway.

• Try it live

End-to-end RAG chatbot that answers questions about my background. Hybrid retrieval combining BM25 keyword search and FAISS dense embeddings with configurable alpha weighting (inspired by my Pinecone internship). Query rewriting layer using Claude for vague-to-specific expansion. Empirical alpha sweep across 15 Q&A pairs using LLM-as-judge found pure semantic retrieval achieved 5.0/5.0 relevance. React frontend with SSE streaming, source attribution, and dark/light mode. Docker containerized, deployed on Railway with rate limiting.

• Try it live

Automated market-making system for CFTC-regulated prediction markets (Polymarket US & Kalshi). OBI microprice estimation, continuous inventory skew, and dynamic spread adjustment. 4-layer risk management from fat-finger checks to system-level circuit breakers. Platform-agnostic adapter pattern so the same engine works across exchanges. Calibration study on 10,000+ resolved markets validated edge before committing capital. Paper trading: +151.9c P&L across 71+ fills in 11 sessions. Runs 24/7 on Mac Mini with PM2 and Discord alerts.

• Check it out

Automated reservation sniper for high-demand restaurants on Tock. Playwright browser automation handling the full booking flow: login, calendar navigation, slot detection with 6-selector fallback, and checkout. Two modes: hourly polling for scanning and precision sniper mode (3-second polls during release windows). Concurrent booking across multiple dates, session cookie persistence, Discord notifications. Deployed 24/7 on Mac Mini via PM2.

• Check it out

This project develops machine learning models for accurately classifying mushrooms as edible or poisonous using a dataset of 3000+ images. Address challenges in mushroom identification and provide valuable tools for mushroom enthusiasts and field scientists.

• Check it out

In this project, we assess the impact of government vaccination policies on Covid-19 spread across countries. Analyze time series data, including infection rates, testing rates, and vaccination policies. Build accurate ARIMAX models to understand policy effectiveness and inform decision-making. Discuss limitations and future considerations for analysis.

• Check it out

This project utilizes physical measurements instead of time-consuming methods, exploring multiple regression techniques to estimate abalone age. With a dataset of 4177 observations and 9 variables, including the response variable "Rings," we address challenges like high correlation and multicollinearity. Our goal is to provide an efficient and accurate approach to age estimation, transforming the abalone industry.

• Check it out

This research project analyzes and classifies Kama, Rosa, and Canadian wheat based on kernel features. Utilizing visualization, correlation analysis, and classification techniques, we uncover distinct characteristics and establish feature relationships. Applying methods like LOOCV and random forest grid search, we provide valuable insights into wheat differentiation and classification.

• Check it out

This project utilizes the power of Shiny and Plotly in R to develop an interactive, real-time dashboard visualizing the COVID-19 impact worldwide. Using the covid19api, the dashboard provides information on daily cases, deaths, and recoveries per country, including an all-encompassing global overview, thus serving as a comprehensive tool for COVID-19 data analysis and insights.

• Check it out

This research project explores how participants' reliance on public predictions can override their own private information in decision-making. Through an experiment involving sequential predictions and payoff incentives, the study investigates information cascades and their impact on rational decision making. The findings shed light on herding behavior and have implications for various domains such as investments, marketing, and voting.

• Check it out