Applied AI, delivered.

The capability behind the work

One builder, the full arc — research to working software.

Every project on this page was taken end to end by one person: Mangethe Zwane, who spent fifteen years in analytical chemistry, sensory science and data analytics for the global brewing industry before turning that rigour on software. The method is the same on every build — start from the real business problem, model the data honestly, then ship working, deployed software. Research → data model → working system, with no hand-off gap where the domain understanding leaks out.

The builds run on a deliberately boring, proven stack — Next.js and React on the front end, Supabase Postgres with row-level security underneath, TypeScript throughout — chosen so that multi-tenant isolation, an immutable audit trail, and 80%+ test coverage are the default rather than an afterthought. AI is used as the primary build assistant to compress the writing of that code, under human review, not bolted into the product: several systems ship with AI optional and off by default.

Where the problem is genuinely hard — turning raw mass-spectrometry data into a scored, cited compound workbook, or standing deterministic statistical evidence behind a beverage Release / Hold / Reject decision — the work goes deeper than wiring an API to a chatbot. It is data engineering and applied AI at R&D depth: cleaning, scoring, ranking, provenance and reproducibility, built to be evaluated by a scientist rather than trusted blindly.

The through-line is range with depth. The same person can architect a compliance SaaS in four days, ship an offline-first consumer PWA within free-tier limits, and coach a plant floor through SDCA/PDCA cadence — because the technical authority and the domain authority are the same authority. That is what this portfolio is built to prove.

Mangethe Zwane · Founder & principal engineer, Khula Platform