At 120,000 employees, the HR intranet stops scaling.
A global aviation group with a six-figure workforce was drowning in policy questions - we built the assistant that gave each employee a personal HR concierge.
A workforce of over 100,000, spanning sixty-plus nationalities and dozens of grade tiers. HR and travel policy entitlements that vary by role, tenure, and grade - and that change often enough to make the intranet outdated within weeks of any update. The result was predictable: intranet searches that returned the wrong policy, email threads waiting for an HR agent to read them, call centers handling questions the policy library already answered. The question that triggered the engagement: could a single assistant deliver personalized, policy-correct answers - in seconds, in multiple languages, at the scale of the entire group?
- 01
Anchor everything to the policy source.
We built the assistant as a RAG system over the official HR and travel policy library, with metadata-aware chunking that preserved the structure policy authors had written into the documents. No paraphrased general knowledge - every answer routes back to a citable policy clause.
- 02
Personalize at the employee level.
The assistant reads the asker's profile - grade, role, tenure, location - and tailors entitlements accordingly. A senior pilot and a graduate engineer asking the same question get different, correct answers. The judgment call: we built guardrails before we built capability. The assistant declines out-of-scope questions explicitly rather than hallucinating an answer.
- 03
Industrialize on the enterprise data platform.
Rather than a bespoke deployment, we shipped the assistant on the group's own enterprise data science platform - the second group-wide GenAI use case to land there. That choice meant governance, monitoring, and lifecycle management came for free.
HR service-center contacts dropped substantially. Employees got instant, consistent, policy-aligned guidance - at any hour, in their preferred language, within three seconds. The HR function recovered capacity that had been absorbed by repetitive policy queries and redeployed it to strategic workforce work. The unlock: a foundation for broader enterprise GenAI rollout, with one production pattern proven inside the group.
“In enterprises, the constraint on GenAI isn't the model - it's governance. The teams that build refusal behavior, profile-awareness, and platform integration first are the ones whose pilots become production.
Sitting on a policy library too dense for self-service and an HR function too valuable to spend on lookups? We help enterprise teams turn knowledge bases into governed AI assistants - at platform scale.
Let's talk
