Built Right.Runs Quiet.

Half a million lines of production code sitting on top of the CRM real estate teams already run. Five machine-learning models score every contact on likelihood to buy, sell, respond or go cold, and write those scores back into fields the agent already looks at. A 242-tool MCP server and a voice agent let the team operate the CRM by chat or speech, and twenty change-data-capture streams feed a BigQuery warehouse where predictions are checked against what actually happened.

• 5 Production ML Models
• 242-Tool MCP Server
• Voice Agent on the CRM
• BigQuery Warehouse

333 articles about Austin bars, restaurants and neighbourhoods, each wired to live civic data rather than opinion. Components embedded in the prose pull a venue's actual health inspection score from the 45,783 records ingested, current Barton Creek flow from USGS gauges, and economic series covering 25 Texas metros and 252 of the state's 254 counties.

• 45,783 Health Inspections
• Live Creek Gauges
• Economy Dashboards
• 333 Articles

Twenty disjoint government and industry feeds resolved into one searchable directory: 188,747 companies, 143,389 consolidated firms, 413,350 unified construction projects and 885,569 work-history records. Statewide licensing data spans all 255 Texas counties. Entity resolution escalates from Postgres trigram matching to LLM disambiguation, and Postgres triggers diff JSONB snapshots to send change alerts only when a project's cost, status or contractor actually moves.

• 188,747 Companies
• 413,350 Projects
• 255 Counties
• Change-Detection Alerts

A self-service property tax analysis platform for Central Texas homeowners. Pulls live data from Travis (TCAD), Williamson (WCAD), and Hays (HCAD) county appraisal districts, scores hundreds of comparable properties using USPAP-standard adjustment methodology, and generates protest-ready evidence packets—complete with personalized ARB hearing scripts, pre-written protest emails, and downloadable PDF evidence. AI-powered insights surface the strongest arguments for each property. A white-label system lets real estate agents offer branded reports to their clients.

• Multi-County Analysis
• AI Protest Insights
• USPAP Adjustments
• White-Label Ready

A flooring installer Austin has trusted since 1984, whose entire back office — every job, builder, installer, work order and invoice — sat with a developer the relationship had broken down with. We recovered the Laravel ERP, migrated it into Sun Tile's own AWS account, and built the CI pipeline it never had, then improved the system in place rather than rewriting a decade of encoded operational knowledge. With the back office owned and stable, we replaced the WordPress storefront on top.

• Hostile-Developer Recovery
• AWS Migration
• CI/CD From Scratch
• Storefront Rebuild

Austin sits on the Balcones Fault, where expansive clay meets Hill Country limestone — and the search results are owned by lead mills that resell a homeowner's phone number to five contractors. TrueLevel wins the click by out-explaining them: 52 engineer-authored content entries, 75 generated suburb pages grounded in each area's actual soil, and a live lookup that takes a street address and chains three public geology APIs to return a real shrink-swell risk classification.

• Address-to-Soil Lookup
• 75 Suburb Pages
• 3 Chained Geo APIs
• Zero-JS Content

A home-accessibility business where the website is the salesperson and a serverless back office runs the follow-up. 15 location pages targeted by 65-and-over population density rather than suburb growth rate, and a deterministic rules engine that maps a 10-question intake against VA, Texas Medicaid, municipal and USDA programs — then emails a personalised funding map inside the same request. No LLM in that path, on purpose: a hallucinated benefit eligibility would do real harm.

• Funding Rules Engine
• Demographic Targeting
• Cron-Driven Follow-Up
• 50 Pages

24,000 lines of dependency-free JavaScript. Evolution Mode subjects the rules themselves to selection pressure — birth and survival conditions mutate, get scored on growth, stability and diversity, and compete as tracked species. A soup-search engine runs 13 discovery strategies, classifies what stabilises by exact oscillator period and spaceship velocity, encodes findings as real apgcodes, and cross-references the global Catagolue database to flag genuinely novel patterns.

• Evolution Mode
• 13 Search Strategies
• Catagolue Cross-Reference
• Zero Dependencies

A WordPress archive of 19,415 visual guides, facts and conversation starters, rebuilt as a Next.js site with no database at all. Content ships as per-category JSON shards, recommendations run entirely in localStorage with a recency-decay engagement model, and shuffling uses a seeded PRNG so a shared link resolves identically for whoever opens it.

• 19,415 Posts
• No Database
• Offline Recommendations
• WordPress Migration

A full-stack lead generation platform connecting homebuyers, sellers, and renters with Austin real estate experts. Features an interactive neighborhood explorer with 80+ areas, multi-step qualification quizzes with readiness scoring, and automated email workflows—driving quality leads through personalized user journeys.

• Interactive Map Explorer
• Homebuyer Quizzes
• 80+ Neighborhoods
• Lead Automation

A mission-driven web application helping Americans discover Social Security benefits they may be eligible for but don't know about. Free 2-minute diagnostic identifies potential benefit sources from retirement, spousal, ex-spousal, survivor, and deceased ex-spouse records. Built with elderly-optimized UX (20px base font, 72px buttons, WCAG AAA compliant). Privacy-first architecture—all processing happens client-side with zero data storage or tracking.

• Multi-Benefit Discovery
• Privacy-First
• Elderly-Optimized
• Client-Side Only

Rescued a beloved South Austin Cajun restaurant from their early-2000s website. The live-music calendar is a Google Sheet staff already know how to edit — read server-side for Schema.org event markup, re-fetched client-side for instant updates, cached to localStorage, and backed by a static fallback so it never renders empty when Sheets is unreachable. Untrusted spreadsheet input is parsed defensively and URLs are allowlisted.

• Sheets-as-CMS
• 3-Tier Fallback
• Live Google Reviews
• Answer-Engine Ready