People ops

Staff ask HR.
HR asks the handbook.
The desk answers both.

Leave accrual, probation rules, disciplinary steps, answered from your approved policy corpus with citations, from your approved policies, off a public chatbot that hallucinates labour law.

$ corvux hr ask --user staff-4421 \
 "Can I carry forward unused annual leave?"

ROLE employee · dept: operations
RETRIEVE Leave_Policy_2026.pdf · §4.2
ANSWER Max 5 days carry-forward · use by 31 March…
CITE Leave_Policy_2026.pdf p.7
✓ no PII sent to foreign LLM · logged

HR in Kenyan enterprises fields the same questions every Monday: annual leave, sick leave proof, probation extensions, remote work eligibility. HRBPs grep PDF folders while staff WhatsApp informal answers that can drift from policy.

The HR policy desk is a private RAG chat over your handbook, collective agreements, and HR circulars, streaming cited answers on Kenya-hosted models. Same muscle memory as ChatGPT; different trust model.

Only what HR has approved.

Employee handbook

Leave, conduct, benefits, grievance, the canonical staff-facing set.

HRBP playbooks

Disciplinary matrices, investigation checklists, restricted to HR roles.

Onboarding packs

Probation, kit allowances, probation review timelines for new hires.

Local addenda

Union agreements, county office variants, tagged by entity.

Cite
Every answer
SSO
Role gates
Audit
Query logs
KE
Host

HR first. Company-wide when citations hold.

01
Corpus audit
Identify conflicting versions, fix docs before indexing, before indexing, ahead of the complaint pile.
02
HRBP pilot
20 HR users hammer edge cases; harvest “couldn’t find” into doc fixes.
03
Staff launch
Self-service for L1 questions; complex cases link to HR ticket form.
04
Govern
Publish workflow + gateway to discourage public ChatGPT on employee data.
A 2,000-person manufacturer reduced “policy clarification” tickets by roughly a third in the pilot quarter, and staff stopped guessing, beyond any story that AI replaced HR, because staff stopped guessing.

Built on the same stack as private ChatGPT, specialised corpus and guardrails for people data. Host on managed API or your GPUs.

HR and legal.

Will it give legal advice on termination?

No. It retrieves policy text and cites sources. Final calls stay with HR and counsel, the desk reduces “where does it say that?” searches.

NSSF, SHIF, PAYE updates?

Only what’s in your published corpus. We re-index when you upload updates, it doesn’t scrape government sites autonomously.

Can managers and staff see the same answers?

Role-based spaces: staff see employee handbook; HRBPs see full policy library including draft guidance if you allow.

Integration with HRIS?

Read-only FAQ layer first. Write actions (leave booking) are separate integrations scoped explicitly.

Pilot the desk with HRBPs

Upload your handbook and top 10 circulars. We stand up a cited chat in two weeks.

Start an HR policy pilot