Case studies
Each case study covers the problem, the architecture decisions and their trade-offs, the delivery pipeline and the project's actual state. Client projects are anonymised unless the client agreed in writing; every figure quoted comes from the code repositories.
See the shared engineering platform →OpenMRS 3, three field Wi-Fi kits, on-site training and a data pipeline up to the reports, for a four-day medical and surgical mission in Cameroon.
Public site and back office on a hexagonal architecture, shipped through a DevSecOps pipeline that produces SBOMs and delivery evidence.
Hexagonal Spring Boot API and practitioner back office: computed slots, temporary holds, confirmation. In production.
One generic Helm chart, validated profiles and Argo CD: several client applications deployed the same way, with no manual commands.
Multilingual site with booking and a conversational assistant, shipped by a pipeline that opens, verifies and merges PRs, with human approval before production.
NestJS modular monolith and Next.js front end, 93 test files and a pipeline that checks HTTP contracts and coverage before any release.
A self-hosted Backend-as-a-Service for client applications: identity, data, storage and messaging behind a minimal configuration.
Next.js site on Kubernetes managed by Argo CD, plus an ongoing migration from a legacy WordPress site that preserves search rankings.
Public site with SEO, accessibility and security headers; the back office was deliberately not built until the need is confirmed.
Showcase site, online quotes and a media back office (WebP conversion, events); structured data for local search.
16 architecture decisions (per-tenant data isolation, anti double-booking ledger, idempotent commands, outbox) and a pipeline that tests cross-tenant isolation.
A deterministic API with a hexagonal architecture: full OpenAPI contract, row-level security and Alembic migrations run as a job.
A design recorded in 16 decisions: modular monolith, readings storage, multi-tenancy, GDPR and data governance, SLOs, fitness-function tests.
A service to manage R2 storage (tokens, PostgreSQL backup and restore), delivered in 10 reviewed phases with 15 architecture decisions.
A real audit of a cluster and its servers: findings, root causes, fixes and an Infrastructure-as-Code modernisation recommendation.
A language model hosted on our own infrastructure, used in production by a client's conversational assistant.
A multilingual site checked by GitHub Actions (lint, typecheck, build, Helm, Docker) and deployed only through Argo CD, with SemVer versioning.