Not mockups or pilots. Every system below is deployed, has real users or real data flowing through it, and is maintained by us. Each case study covers the problem, the architecture, and the parts that were genuinely hard.
An AI operating layer for real estate teams, on top of Follow Up Boss
Half a million lines of production code: 242 MCP tools, a voice agent that acts on the CRM, five machine-learning models retrained behind a promotion guard, and a BigQuery warehouse fed by twenty change-data-capture streams.
Texas construction intelligence, assembled from twenty public data sources
188,747 companies, 413,350 unified construction projects and 885,569 work-history records, resolved out of twenty disjoint government and industry feeds into one searchable directory — with statewide TDLR coverage across all 255 Texas counties.
A city guide where every recommendation is backed by public data
333 articles about Austin bars, restaurants and neighbourhoods, each wired to live first-party data — 45,783 health inspections, real-time Greenbelt creek gauges, and county-level economics for 252 of Texas's 254 counties.
Foundation repair answers for Austin homeowners, from an engineer
A referral platform that beats the foundation-repair lead mills by out-explaining them — 52 engineer-authored content entries, 75 generated suburb pages, and a live address-to-soil-risk lookup chaining three public geology APIs.
A home-accessibility business, built as software first
50 pages of aging-in-place content targeted by 65-and-over population density, plus a deterministic funding-eligibility engine that maps a 10-question intake against VA, Medicaid and municipal programs and emails the answer before the visitor closes the tab.
Rescuing an ERP from a hostile developer, then rebuilding the storefront
A 40-year Austin flooring company whose entire back office sat with a developer who had turned hostile. We got the ERP out, moved it into their own AWS account, built the CI pipeline it never had — then replaced the WordPress storefront on top.
19,415 posts, no database, one very stubborn serverless bundle
A WordPress archive of 19,415 posts rebuilt as a static Next.js site with zero database — including a localStorage-only recommendation engine and a seeded PRNG so 'random' content stays reproducible.
A South Austin Cajun institution, off a 2000s website
A restaurant site rebuilt on Astro with a live-music calendar that staff edit in a Google Sheet — backed by a three-tier fallback chain so the calendar never goes blank, even when Sheets does.
Conway's Game of Life, with rules that evolve and patterns it discovers itself
24,000 lines of dependency-free JavaScript: an Evolution Mode where rulesets mutate under fitness pressure, and a soup-search engine that classifies its discoveries as real apgcodes and checks them against the global Catagolue database.
24K lines, zero dependencies
13 soup-search strategies
23 patterns in the library
Read the case study
Also shipped
The rest of the portfolio
Smaller builds, side projects, and sites that don't need a full case study — but are live, maintained, and doing their job.
Self-service property tax protest analysis for Central Texas homeowners. Pulls live data from the Travis, Williamson and Hays appraisal districts, scores comparable properties using USPAP-standard adjustments, and generates protest-ready evidence packets with hearing scripts and PDF exhibits. White-label mode lets agents deliver branded reports.
Anonymous one-to-one video chat with interest matching and country filters, free and account-free. Two Cloudflare Workers handle matchmaking and WebRTC signalling at the edge, backed by a 26-table Postgres schema covering moderation, safety reporting and payments.
A free two-minute diagnostic that surfaces Social Security benefits people are eligible for and don't know about — retirement, spousal, ex-spousal, survivor and deceased-ex-spouse records. Built for older users at WCAG AAA, and computed entirely client-side so no personal answer ever leaves the browser.
A lead platform matching buyers, sellers and renters with Austin agents. An interactive explorer covers 80+ neighbourhoods, and multi-step qualification quizzes score readiness before a lead reaches an agent's inbox.