Projects

ML model predicting ownership changes from land tax record datasets. Predecessor to Ranger's data layer — trained on classification targets with heavy feature engineering and class imbalance handling.

• →Ownership change prediction from county land tax datasets
• →Class imbalance handling — oversampling and weighted loss
• →Feature engineering on county assessor data
• →Tabular model architecture — predecessor to Ranger's data layer

PDF table parser for Texas W-1 drilling permit forms from the Railroad Commission. Extracts structured permit metadata — operator name, legal description, formation, API number — from unstructured PDF tables using regex pipelines.

• →PDF table parsing from RRC W-1 permit forms
• →Regex extraction of operator, formation, and API number fields
• →DataFrame output with typed columns
• →Prototype for Ranger's OCR ingestion pipeline

Student collaboration and Q&A platform designed to bridge the gap between coursework and applied problem-solving. Combines computer science and business thinking to build a knowledge-sharing environment for university students.

• →Won university pitch competition
• →Combined CS and business domain thinking
• →Designed for student collaboration workflows

Applied machine learning research across financial fraud detection and loan default prediction. Trained and evaluated models across 8+ architectures using real-world financial datasets with heavy feature engineering and class imbalance handling.

• →Financial fraud detection (IEEE-CIS dataset)
• →Loan default prediction (LendingClub dataset)
• →Class imbalance handling — focal loss, class weights
• →Evaluation across 8+ model architectures

Data collection pipeline for Texas mineral rights research. Playwright-based county clerk record scraping, PDF ingestion and OCR parsing via pdfplumber, and Texas Railroad Commission dataset processing — the source layer for Ranger Discovery's document corpus.

• →Playwright-based county clerk portal scraping
• →PDF ingestion and OCR parsing via pdfplumber
• →Texas Railroad Commission dataset processing and cleaning
• →Document corpus pipeline feeding Ranger Discovery

A living fantasy world built from first principles — tectonic plates and atmospheric science before nations, nations before politics, politics before mythology. QGIS-modeled geography, a complete pantheon, a magic system grounded in planar physics, and the Arkadim: an ancient AI that merged with a human consciousness and went mad.

• →QGIS layer structuring — tectonic plates → climate → hydrology → biomes
• →Full creation mythology with emergent pantheon — gods ascend through mastery or necessity
• →Magic system grounded in planar physics — Limbo, madness mechanics, entity mediation
• →The Arkadim — ancient AI/human hybrid driven mad by grief, reanimates dead as flesh-machine constructs
• →Nations and political systems derived from geography — not placed arbitrarily

Custom TTRPG system designed from scratch — mana economy, probability-balanced combat, leveling architecture, and skill trees. Built from systems design principles: balanced mechanics, emergent complexity, no reliance on existing rulesets.

• →Mana economy with resource constraints and regeneration logic
• →Probability-balanced combat resolution system
• →Leveling architecture and skill tree design
• →Custom ruleset — no existing system dependencies

Long-form D&D 5e campaign architecture with political factions, branching consequences, and player agency. Rival factions with independent motivations, consequence branches, and a setting where geography shapes economics and power.

• →Political factions with independent motivations and goals
• →Branching consequence architecture driven by player choices
• →Long-term player agency design
• →Geography-driven political and economic systems