Andrew Erbs

A progressive BJJ training application that accelerates skill acquisition through spaced repetition (SM-2 algorithm) and evidence-based mental visualization techniques. Combines physical training logs, guided 5-minute visualization sessions, smart scheduling, research protocols (nutrition, cardio, strength), and comprehensive progress tracking.

A comprehensive analytical framework for evaluating NFL quarterback performance and contract decisions through advanced statistical modeling and era-adjusted metrics. Combines systematic data collection from Pro Football Reference with ridge regression analysis to reveal organizational biases in quarterback evaluation and predict contract outcomes.

An intelligent project management automation system that integrates Jira with Claude AI to automatically analyze requirements, generate implementations, and evaluate deliverables, reducing manual effort by 70-90% while maintaining high quality and consistency.

An automated news aggregation and analysis system that uses Claude AI to generate targeted search queries, fetches articles from NewsAPI, synthesizes summaries, and delivers reports via GitHub Issues with intelligent caching.

An interpretable pre-fight MMA handicapping pipeline that estimates a calibrated six-outcome probability distribution (win/lose by finish or decision) from tiered fight and fighter data. Combines cross-promotion ELO construction, ELO-weighted style features, matchup interactions, multinomial logistic regression, and explicit confidence intervals—using only information available before the opening bell.

A research-oriented pipeline that models the Enron corpus as a communication graph and aims to surface behaviorally unusual actors from structure and metadata alone (headers, timing, threading, CC patterns)—not email body content. The design uses person-level nodes, a directed multigraph, rolling temporal windows with per-person baselines, and self-supervised graph autoencoders with optional temporal drift; evaluation is planned around held-out indicted executives and event-aligned sanity checks. The repo currently ships an installable Python package scaffold, documented architecture decisions, corpus extraction from the public maildir tarball, a root-level runner with per-stage rebuild vs cache flags, tests, and a plain-language roadmap for parsing, graph build, features, training, and eval.

Cross-city market analysis of short-term rental investment opportunities across 22 US cities, combining Airbnb listing data with Zillow real estate prices. Identifies profitable property segments, competitive landscape dynamics, and market entry barriers through statistical modeling and market professionalization metrics.

An AI-powered application that analyzes jiu-jitsu techniques and generates personalized training strategies, flow charts, adversary response plans, and analytical insights for practitioners of all levels.

A robust web scraper application that extracts Penn State football schedules and creates standardized calendar files, allowing fans to automatically sync game information with their personal calendars.

A Python web scraper that automatically extracts Yale football schedule information and generates an iCalendar file for easy integration with calendar applications.