Skip to content
OL, Otávio Lima — home
All work

Case study

Modernising an enterprise digital-twin platform

60% faster, and the architecture became the company's reference.

Role
Senior Software Engineer, Bentley Systems
Period
Mar 2023 — Present
Stack
  • React 18
  • Zustand
  • React Query
  • Module Federation
  • Azure DevOps
  • Playwright
Client work — code and screenshots withheld

TL;DR

  • Migrated a large React application from a Redux store to React 18 with Zustand and React Query, cutting page performance cost by 60%.
  • Replaced iframe integrations with six independently deployable micro-frontends on Module Federation.
  • Built the Azure DevOps pipeline to Azure CDN that became the company-wide reference for frontend delivery.

Key figures

Faster
60%
Micro-frontends replacing iframes
6

The problem

An enterprise digital-twin platform had grown into a large React application with two problems that made each other worse.

The first was state. A single Redux store held almost everything, and components subscribed to far more of it than they read, so ordinary interactions re-rendered large parts of the page.

The second was integration. Features owned by other teams arrived as iframes. Each one was a separate document with its own load, its own copy of everything, no shared styling, and a postMessage boundary for every conversation with the page around it.

The product stays unnamed here, and the diagram above is redrawn from scratch: none of this work's code or UI can be shown.

Decisions and trade-offs

React Query + ZustandoverOne Redux store

Most of what lived in the store was server data, so the store was doing a cache's job without a cache's tools. Splitting state by kind fixed both problems at once: server data moved to React Query, which caches, deduplicates and refetches it, and the small amount of genuine client state moved to Zustand, where a component subscribes to exactly the slice it reads. The migration to React 18 landed alongside it.

Module FederationoverIframes

Iframes isolate by building a wall. Federated modules keep the isolation that matters — each team still builds and deploys on its own — while sharing the page, the React instance and the design language. Six iframe integrations became six independently deployable micro-frontends.

One reusable pipelineoverA pipeline per team

Six independently deployable apps means six deploy paths, and six chances for them to drift apart. I built one reusable Azure DevOps pipeline that takes any of them to Azure CDN the same way. It became the company-wide reference for frontend delivery.

A testing pyramidoverEnd-to-end tests alone

Independent deploys are only safe if each piece can be trusted on its own. Jest and React Testing Library cover behaviour quickly and close to the code; Cypress and Playwright sit on top for the journeys that cross boundaries.

Results

  • 60% lower page performance cost after the state migration.
  • Six micro-frontends in place of iframe integrations, each deployed on its own schedule.
  • A delivery pipeline that the rest of the company now builds on.
  • Page performance cost100 → 40-60%
Indexed: before the migration = 100. The figure is the published 60% reduction, not an absolute timing.

How it was measured

The 60% comes from the same two sources before and after the migration: the platform's observability data and Lighthouse reports. Together they tracked the three things the state migration was meant to change — how fast pages loaded, how many re-renders an interaction caused that it did not need, and how many duplicate requests went out for data the page already had.