Selected work · Proposal, not yet approved

Planning the rebuild of an internal staff and asset system

A review of the .NET 6 system HR and IT use to run staff and company assets, a weighted comparison of seven ways to rebuild it, and a phased plan for staff sign-in that changes nothing people see until it has to.

Role
Reviewer and author of the rebuild plan
Company
Machines
When
September 2026
  • .NET 6 today
  • .NET 10 target
  • Blazor
  • Hangfire
  • OpenIddict
The seven rebuild options by weighted score. The recommendation is highlighted.
lines of C# reviewed
~32k
rebuild options scored on weighted criteria
7
sign-in phases, each a safe place to stop
5

Problem

HR and IT at Machines use an internal system to onboard staff, grant and revoke access to other systems, assign devices, keep asset registers, and offboard people when their last day arrives. It runs on .NET 6, which left support in 2024, as three separately deployed parts over one database. The team needed to decide whether to patch it, rebuild it or replace parts of it, and how to bring staff sign-in in line with the company’s other staff systems at the same time.

Constraints

  • Security and maintainability first. Options were scored on security (35%), maintainability (30%), risk (20%) and flexibility (15%), and any option with the lowest security score was ruled out.
  • Target .NET 10 LTS, not .NET 8, whose support ends in November 2026.
  • A new UI was acceptable if it made the system easier to maintain. Flutter was ruled out for this system.
  • Nobody’s sign-in could change without a reason they would accept. Outlets and departments share accounts, and not everyone has a Microsoft account, so a single Microsoft sign-in could not simply be switched on.

What I decided and built

I read the code end to end and wrote an architecture review, then scored seven ways to rebuild it, from patching what exists to buying part of it. The recommendation is one application instead of three: a modular monolith on .NET 10, organised by feature, with a Blazor UI, Hangfire for scheduled work, and Linux behind nginx. Keeping three separate parts multiplies deployments, sign-in paths and failure modes without buying independent scaling, because the web front end is the only real client of its own API.

For sign-in, the plan has five phases, and the first two change nothing anyone sees: security and shared login code move behind the existing screens. Small opt-in changes, such as authenticator codes for administrators, come next. Phase 3 is a dedicated login service on .NET and OpenIddict that keeps the same login forms and gives each person one password. Single sign-on is a final, optional step. Each phase is a safe place to stop.

I also wrote up Keycloak as an alternative, including what it would mean for keeping the existing login screens, so the team could compare a product with a build.

Outcome

The team received three short documents: an architecture plan (8 pages), a sign-in standardisation plan (7 pages) and the Keycloak alternative (6 pages). The rebuild is waiting for approval, so this is a plan and not a shipped system.

What I’d do differently

I scored the first version of the comparison on migration risk alone, which flattered the existing system for not moving. The corrected version also counts the risk of running each result. Next time I would agree the risk measure with the team before scoring anything.