No invented metrics
When an exact figure is confidential or unavailable, I describe the technical change instead of manufacturing a percentage.
Engineering case studies / 01—09
Engineering decisions, shown in context.
Nine detailed case studies: three anonymised professional systems and six open-source Labs projects. Each one explains the constraints, the decisions, what was built and where the evidence stops.
00 / NOTE
P01 P02 P03
Work worth explaining.
A stack list tells you very little. This archive separates professional work from Labs projects, then follows the decisions, verification and limits behind each system.
Decisions before tools
The technology matters, but only after the problem and the trade-off are clear.
Names withheld, work intact
Organisations and commercial details are omitted. The engineering scope is kept specific.
Case-study archive09 CASES
9 case studies
CASE / 01
AI workflow platform cloud migration.
In roughly one month, I moved a multi-component document and AI-workflow platform from an environment-specific on-premise setup to reusable Terraform, GKE and a controlled blue/red delivery path on Google Cloud.
CASE / 02
Archival workflow management.
Over roughly three months, I built a Spring Boot service and React Single-SPA microfrontend around Camunda so operators could create archival sessions, move source packages through a nine-step lifecycle and track the resulting output.
CASE / 03
Retail ERP maintenance and evolution.
Across approximately nineteen months, I worked through SQL Server, .NET Framework 4.8 and a KnockoutJS frontend in a live retail ERP, covering features, defects, reporting, two courier integrations and stock-receipt modernisation.
CASE / 04
CareerOS Local.
CareerOS Local combines a Tauri desktop shell, a FastAPI sidecar, a versioned SQLite vault and a required local LLM runtime. The result is a career utility that keeps source facts, documents and analysis on the user’s device.
CASE / 05
ELIZA Lab.
ELIZA Lab is a Rust machine-learning pipeline for training, calibrating and inspecting an intent classifier locally. It replaces a misleading therapy-bot premise with a reproducible, non-clinical experiment that can abstain.
CASE / 06
DjenisAiAgent.
DjenisAiAgent observes a Windows or browser interface, requests one structured Gemini action, checks it against runtime permissions and feeds the verified result into the next turn.
CASE / 07
DIG.
DIG pairs a real terminal Gopher client with a deterministic browser explorer. The CLI opens gopher:// resources over bounded TCP; the web edition teaches the same parser through a safe fixture.
CASE / 08
IntegraDraw.
IntegraDraw is a Java desktop and TypeScript Canvas workbench for comparing midpoint and trapezoidal sums with a Simpson reference. Both runtimes share versioned numerical cases and explicit tolerances.
CASE / 09
VECTOR.
VECTOR turns a fictional placement cohort into a browser-only operations board. Search, status filters, progress views and milestone updates run locally with no account, backend or analytics service.
00 / 00
No case study matches those filters.
Try a broader phrase, another collection or every topic.
Have a system that needs a clearer path forward?
Bring the difficult part. We can work through the architecture, delivery plan and first useful release together.