Jan 2026Case studyIn production1 min read

Flutter web admin migration with no downtime

Every admin page had its own upload, validation and error handling, and closing a dialog could wipe a half-finished mapping. I replaced them with one import flow that checks files in the browser and protects unsaved work, moved over while ops kept using the tools.

Every admin page parsed uploads its own way, and one stray click could wipe a half-built mapping. I moved them onto one checked import flow, module by module, without taking the tools offline.

Challenge
Every admin page in the Flutter web app had its own upload, mapping and validation code, and closing a dialog could wipe a half-finished mapping.
Approach
Shared browser-side sheet validation, one import wizard and an unsaved-changes guard, moved over one module at a time while ops kept working.
Impact
Core modules moved with no scheduled downtime, and the integration dialog now carries 7 CRM connectors: Zoho, Salesforce, HubSpot, Facebook Lead Ads, WhatsApp Business, S3 and GCS.
One import flow for every admin page
  1. 1Upload

    Ops drop in a CSV or XLSX file.

  2. 2Validate in the browser

    Columns, types and lengths are checked before anything is sent, with errors shown on the exact row and column.

  3. 3Map in one wizard

    Every import walks through the same stepper, with the org's roles and processes loaded once.

  4. 4Guard unsaved work

    Closing the dialog with unsaved mappings asks first instead of wiping them.

  5. 5Send in batches

    Only a valid, confirmed mapping is sent, as one batched request.

7

Integrations I built the setup for

  • Zoho
  • Salesforce
  • HubSpot
  • Facebook Lead Ads
  • WhatsApp Business
  • AWS S3
  • Google Cloud Storage

One integration dialog

Connection, field mapping and validation for each, on one shared dialog instead of one flow per admin page.
Integrations on the shared flow
7: Zoho, Salesforce, HubSpot, Facebook Lead Ads, WhatsApp Business, S3, GCS
Scheduled downtime
None: modules moved over one at a time

Problem

Admin views accumulated three structural issues:

  1. Duplicated ingestion plumbing: parse, map, validate, and error UI rewritten per page
  2. Silent form loss: complex mappings vanished on accidental modal dismiss
  3. Fetch churn: wizards re-queried org metadata on every step

Constraints

  • Ops had to keep using the tools during migration
  • Large sheets needed fast client-side pre-validation
  • Validation contracts had to stay aligned with existing CRM schemas
  • Scope is the Flutter-web admin surface specifically. Runo also runs a separate, independently-managed Vue.js admin module untouched by this migration
  • Bulk customer/allocation mutations run through AWS Lambda rather than a monolith endpoint, for cost isolation and reliability under spiky admin traffic

Approach

Diagram
MismatchValidMutatedCommitCSV / XLSXUpload ViewSheet ok?Inline Row/Column ErrorsCustomStepperMapping PreviewUnsavededits?Unsaved WorkConfirmationBatched API Request

Patterns

  1. SheetValidationUtil: column presence, types, length bounds in-browser
  2. wrapDialogWithDirtyCheck: intercept dismiss/navigation with unsaved edits
  3. CustomStepper: one wizard shell for ingestion flows
  4. Normalized domain cache: roles/processes loaded once per flow

Outcomes

  • Core modules moved without a scheduled admin blackout
  • Accidental dismissals stopped wiping multi-field mappings
  • Later CRM integration UIs reused the same upload/validation contracts
  • API key management for downstream integrations (Meta, WhatsApp Business API, HubSpot, Zoho, Salesforce) lives on this surface, inheriting the same primitives

The lasting change: a new admin flow or connector now starts from the shared import, validation and setup flow instead of being rebuilt page by page.