55 rules. 200k records.
Daily. In production.
An enterprise SAP S/4HANA migration needed live master-data validation — no vendor tool covered the business's specific rule logic, and nothing caught bad records before they hit the new system.
A Python and Streamlit data quality engine — 55 custom rules running daily against live master data, scoring records and routing every issue to a business owner, with a GPT-powered layer that turns failed-rule patterns into a plain-English root-cause report.
171,000 issues caught before go-live, across 993 monitored columns. No vendor. No off-the-shelf tooling. Built, deployed and running.
Full order lifecycle,
from brief to delivery.
A manufacturer of premium wellness equipment had no way to show customers real progress on R50,000–R250,000+ custom orders — every update was a phone call or email.
BuildTrack — a magic-link customer portal, no app or password. Nine build stages, workshop photos approved before customers see them, live financials, and QR-code field uploads for workshop staff.
6–14 week builds, fully visible to the customer. Support calls during the build phase dropped to zero.
One codebase,
three clubs, 15 countries.
Open water swimmers in South Africa had no shared source for real-time water conditions or hazards, and no way to coordinate group swims or prove a qualifying crossing actually happened.
LoadKit — SwimLoading. Live conditions, GPS swim logging, group coordination, challenges and leaderboards, and an Evidence Tracker that captures qualifying-swim data — duration, water temp, GPS, observer contact — the moment it's logged, instead of an observer reconstructing it by hand afterward.
Self-funded. Zero marketing. Built and grown solo. Maurten sponsorship followed.
"Dave Welensky verifying my day conditions with an unbelievable app — check it out swimmers."
Yolanda Carstens — open water swimmer, SA community
Coordinating boats and crew,
safely, from a phone at sea.
A Cape Town ocean-swim business was coordinating boats, crew and swimmers by phone and spreadsheet — no shared view of who was where, and no way for a family to know if a swimmer was safe mid-swim.
A real-time operations platform — mission scheduling around weather windows, 11 incident types tracked, automatic alerts if a boat goes quiet, and a live tracking page that goes well beyond a boat blip on a map: position against the planned route, pace and stroke-rate charts, water temperature, and a running check-in feed from the crew.
Live since July 2026. Built for cold, wet gloves on a moving deck — not office UX.
A whole ops platform,
built and deployed solo.
A consulting firm was running projects, timesheets, leave and expenses across spreadsheets, email chains and disconnected tools — nothing talked to anything else.
A 14-module internal operations platform — timesheets with invoicing-grade billing rules and automated reminders, projects, leave, expenses, skills and staffing search, onboarding and assessments. Role-based access, audit-logged, mobile-ready.
53 page routes, 44 API endpoints, 51 database tables — built and deployed solo.
Rebuilt for how AI
answer engines read a site.
A professional-services firm's WordPress site was slow, hard to maintain, and invisible to how AI answer engines actually read a page — and had just been hit by a mail-spoofing incident on the old host.
A 1:1 migration to Next.js — same 22 pages, same brand, same copy voice — rebuilt with schema.org markup, an llms.txt file, explicit access for GPTBot, ClaudeBot and PerplexityBot, and hardened SPF/DMARC.
Same site, rebuilt for speed and machine readability. This page uses the same approach.
Turning decades of
investment judgment into a system.
Two investment domain experts had decades of valuation judgment locked in spreadsheets and memory — nothing evidence-based, nothing continuously running.
An equity-market valuation platform combining market and company data, historical PE valuation ranges back to 1960, forward earnings expectations and analyst consensus into one evidence-based system — every conclusion traceable to its underlying data.
3,200 PE observations across 4 indices, in private access with domain experts, ahead of a licensed live market-data integration.