All Shipped by HR

Shipped by HR

Building an ER Case Management System at Work.Life

Olya Yakzhina · Work.Life

In this episode Olya Yakzhina returns to show us yet another incredible build; HR.Life, an internal app created with Lovable & Claude to help managers submit, track and manage ER cases end-to-end, supported by the People team.

What they built

  • An internal web app called HR.live - an end-to-end ER case management system for managers at Work.Life
  • Covers three case tracks: capability, probation, and disciplinary
  • Guides managers step-by-step through formal processes without needing to go back and forth with the People team
  • Key features include:
    • Intake form capturing employee details, start date, and case track
    • Informal conversation log with a built-in guardrail (blocks progression if informal conversations happened fewer than two weeks prior)
    • Auto-generated meeting invitations using compliant, legally required language
    • AI-generated meeting scripts tailored to the specific case and concerns raised
    • Outcome logging (including wellbeing/health concern prompts) and outcome letter generation
    • Appeal and gross misconduct investigation report generation
    • Full case folder with exportable records
    • Admin dashboard for the People team with activity trail, case oversight, and DPO-ready data governance statement
    • Slack alerts to the People team when a new case is raised or review is needed
    • Direct calendar booking link for manager–People team pre-meeting check-ins
    • Real names anonymised before being sent to AI (initials only)

Why they built it

  • Upcoming UK employment law changes (day-one rights moving from two years to six months of service) raised the compliance stakes for every ER process
  • Inconsistent quality and depth of ER handling depending on the People team member's headspace or bandwidth
  • Managers were passively handing cases to HR rather than owning them
  • Staff were using personal ChatGPT accounts to draft ER-related content, creating data leakage risk
  • Need for better data governance and a single, auditable case record
  • Desire to free up People team time for the genuinely human, judgment-heavy moments

Tools used

  • Claude (company account) - used to write ~99% of the application code
  • Lovable - used to deploy the app; code exported as a zip from Claude and imported into Lovable; Lovable credits used only for bug fixes and tweaks
  • Slack - integrated for real-time alerts to the People team when cases are raised or need review
  • Internal HR policies (disciplinary, grievance, appeal, capability, probation) - uploaded as source logic to constrain AI outputs
  • Existing HR templates (invitation letters, meeting scripts, outcome letters) - fed in as drafts so the AI generates from presets rather than hallucinating

How you could build this yourself

  1. Audit your existing ER documents - gather all policies (disciplinary, capability, probation, grievance, appeal) and every template you currently use (invitation letters, scripts, outcome letters). These become your guardrails.
  2. Map the full case journey for each track before you write a line of code - identify every decision point, required waiting period, and compliance checkpoint (e.g. five working days' notice for formal meetings).
  3. Open Claude (or a comparable LLM with a company account) and use it to build your app logic step by step, uploading your policies and templates so outputs are grounded in your actual documents, not generic AI responses.
  4. Export your code as a zip and import it into Lovable to deploy the front-end; use Lovable credits only for testing and bug fixes to preserve your allowance.
  5. Build in hard guardrails, not just soft prompts - e.g. block progression if informal conversations are too recent; block invitation sending before People team approval.
  6. Add a human-in-the-loop checkpoint at every high-stakes moment: invitation review, script sign-off, outcome letter approval. Wire Slack alerts so the People team is notified instantly.
  7. Protect data - password-protect access, restrict logins to relevant roles, anonymise real names before they reach the AI layer (use initials only), and generate a DPO-ready data processing statement from the outset.
  8. Pilot with one case track (e.g. capability) before building out disciplinary and appeals, so you can test compliance and usability with real managers first.
  9. Book a direct calendar link between managers and their People team partner into the flow - this ensures human conversation happens at the right moment without requiring the manager to chase.

Source

Built by Olya Yakzhina, Head of People at Work.Life, and presented as part of an Open Org "Shipped by HR" session.