YUKEN
Selected work · United Nations

Modernising a mission-critical platform with AI assisted development

Inside a United Nations team, we rebuilt a core module of a mission-critical planning platform with Claude Code, trained the team to work the same way and handed over everything needed to carry it on. Six weeks from first call to working solution.

Client

United Nations

Year

2026

Engines

Solutions + Skilling

Engagement

Six weeks, advisory

The team runs a planning and performance platform used across United Nations peace operations. It worked, but it had barely changed in years. A full redesign existed on paper and had never shipped, the performance problems were well known and there was no internal full-stack capability to take any of it on.

What we did

Build the system, skill the team

One engagement, both engines. The rebuild proved the workflow, then the workflow became the training.

01

Understand the legacy

The platform had run largely unchanged for years. We unpacked the application source, restored a copy of its database into a safe local environment and mapped how the whole system actually worked, so every change that followed could be tested against real structures without touching live data.

02

Rebuild the pilot module

We rebuilt the platform's recommendations module from scratch as a modern web application, React and TypeScript running inside the organisation's existing Power Platform estate, reading and writing the same Azure SQL data through a clean stored procedure layer. A redesign that had sat unimplemented for years finally shipped, matched to the design file exactly. The build was published and verified with 16 of 16 automated end-to-end checks, and a rebuilt version of the original app was kept as a guaranteed fallback.

03

Run the whole build through Claude Code

The entire rebuild was delivered with Claude Code. We set up an agentic workflow in which the assistant edits the code, deploys it and verifies its own work against the running application, with automated browser tests capturing what a user would actually see. One practitioner delivered in weeks what would conventionally have needed a project team.

04

Train the team to work the same way

We designed and delivered live training on agentic AI assisted development, built around the exact workflow used on the rebuild. A full slide deck, an instructor guide, a hands-on practice repository and working guardrails: plan before you build, keep secrets and real data out, verify before you ship. Claude Code was the reference tool, with the alternatives covered properly so the team's procurement options stayed open.

05

Hand everything over

The engagement closed with a complete enablement pack: technical documentation for the new module, a step-by-step deployment runbook, a reproducible Claude Code workflow guide, a reusable project-memory template and recommendations for governance and safe AI use in development. Everything the team needs to keep building without us in the room.

Outcomes

What the team was left with

  • Core module rebuilt as a modern web app and verified end to end, 16 of 16 automated checks passing
  • A redesign that had waited years on paper implemented exactly
  • Team trained live on agentic AI assisted development with working guardrails
  • Deployment into the organisation's own environment prepared and unblocked
  • Full handover pack covering documentation, deployment, workflow and governance
Stack

Claude Code · React · TypeScript · Power Apps · Azure SQL · Power Platform CLI · Playwright · MCP

Start here

Got a legacy system and a team who need to own it?

This is the shape of engagement we like best: build the thing, prove the workflow, then leave the capability behind. Tell us where you are and we will scope it.