The project records are modular and shared with the skills page, so skill associations, dates, descriptions, and proficiency metadata live in one place.
Shows the full skill set without role-based filtering so the broader technical range stays visible.
A published Python library for classification and regression that learns readable, multi-way numeric splits. Rather than representing a measurement as a chain of binary thresholds, RangeBranch can express meaningful operating ranges directly in a single decision-tree node.
A food-network analysis that parsed roughly four million recipe records, extracted ingredients and spices, and built connected views of how spices relate to one another, to broader ingredients, and to cuisines.
A full-stack dashboard builder where users arrange charts, KPIs, tables, and interactive controls into the dashboard they want, then an agent builds the supporting pipeline from both ends: the requested visual outputs and parameters on one side, and approved data schemas, business rules, and source lineage on the other.
A web application that turns an origin point and a user-drawn bounding box into polished isochrone posters or vertical animations, making advanced accessibility visualization easier to configure and repeat.
A reproducible data-science pipeline that reconciles a city-by-city legal directory snapshot into firm, practice, school, and recognition-cohort intelligence with uncertainty-aware rankings and held-out validation.
A desktop blend-design lab that creates an initial spice blend from a user idea, then passes it through a cyclic panel of persona-driven judges who critique and refine the formula into a stronger final blend.
A historical discovery platform for exploring who was alive at the same time, how influential figures connect across eras, and how people cluster by period, category, location, and relationship type.
A collaborative planning-poker product built for a department to run estimation sessions, collect story-point votes, and use historical work to recommend estimates for new assignments.
A full-stack application for schools, clubs, and organizations to collect personalized messages, configure visual templates, and generate polished print-ready materials at campaign scale.
A searchable dermatology reference designed to make clinical terms easier to browse, compare, and study through a focused web experience.
A full-stack coworking platform that gives operators a visual way to configure office spaces, manage resources and bookings, and deliver a branded member experience that can grow into a white-label product.
A long-horizon simulation project that models AI consumers and AI companies to test pricing, product decisions, expenses, purchases, and bankruptcies when real-world data is incomplete or unavailable.
A recurring scraper that collects atmospheric measurements across the continental United States using image processing, interpolation, and scheduled data capture to create a dataset that did not already exist in usable form.
A set of end-to-end automation workflows that detected orders, collected customer data, selected assets, resized images, and routed deliverables into print and fulfillment systems with minimal manual handling.
A content-generation system that scraped athlete names across leagues, ranked them by recency and prominence, arranged them into team-logo compositions, and connected the outputs to an automated storefront.
A workflow for transcribing and analyzing large volumes of call recordings so that business teams could inspect common issues, trends, and operational friction at scale.
A retrieval-oriented workflow that gave teams faster access to SOP content and internal operating knowledge without forcing them to dig through documents manually.
A map-generation system that used QGIS and OpenStreetMap data to create thousands of product variations across locations, colors, and print-ready outputs.
A machine-learning exploration engine that maps molecules, taste descriptors, ingredients, and recipes into a searchable semantic space for discovering flavor relationships.
A lightweight United States routing engine that estimates ZIP-to-ZIP paths by building a nearest-neighbor graph over geocoded ZIP nodes and searching it with A*.
A memory-conscious road-network router that snaps ZIP codes to a North American road graph, estimates travel time by road class, and produces route and accessibility maps.