I build practical web products, reliable AI features, and the unglamorous infrastructure that lets them keep working after launch.
See selected work01 / Selected work
Each project starts with a real point of friction, then gets shaped around clear constraints, useful feedback, and a path to production.
A shareable year in GitHub activity, generated from a single username.
It turns raw contribution history into a durable report with useful stats, generated commentary, and a deliberately optional roast.
Engineering focusPer-user/year caching, idempotent report generation, and an AI fallback path.
A better answer to: which repositories should I show for this role?
The tool reads a role description against repository metadata, ranks the evidence, and returns points that can actually inform an application.
Engineering focusGitHub and AI orchestration, ranking prompt contracts, and structured response shaping.
A small domain used to test dependable patterns for LLM-backed features.
A reference build for schema-checked output, cache-first responses, and local fallback data when an external service is having a day.
Engineering focusRuntime schema guards, cache strategy for token cost, and failure-tolerant service boundaries.
Workout tracking split across a client app and a dedicated API service.
An exercise in choosing boundaries early: a focused interface, an Express domain layer, and independent frontend and API deployments.
Engineering focusService separation, REST resource design, and independent deployment workflows.
02 / How I work
I like the part where an unclear problem becomes a useful system: asking sharper questions, making the smallest honest version, and improving it with evidence.
React, Next.js, Tailwind CSS, Vite
Node.js, Express, REST APIs, authentication
PostgreSQL, MongoDB, relational and document modelling
TypeScript, GitHub Actions, Linux, deployment
OpenAI API, structured outputs, practical evaluation
03 / The record
Umuzi
Umuzi Experience Lab
Umuzi Academy