01 — Problem
The friction
Repetitive HR questions clog tickets and slow down onboarding.
HR Buddy
HR Buddy is the always-on colleague for people ops. New hires get guided onboarding, employees get instant policy clarity, and HR reclaims time for high-touch conversations.
- Ticket volume
- −40%
- Time-to-answer
- < 30s
- Onboarding CSAT
- +18
Demo
See it in motion
A short walkthrough of the agent loop — from signal in to decision out.
Demo reel
Onboarding day-one walkthrough
Overview
Case at a glance
Target users: People ops, managers, and every employee with a question.
02 — Solution
The approach
A conversational HR agent grounded in company policies and onboarding playbooks, with safe escalation whenever an answer needs human judgment.
03 — Value
Why it matters
Instant clarity on policies and a smoother first month for new hires.
How it works
The agent loop
From raw signal to a decision HR can defend.
- 01
Understand
Interprets the question, resolves the employee context, and decides whether policy or process applies.
- 02
Ground
Retrieves the exact policy passages and playbook steps, scoring retrieval confidence before answering.
- 03
Answer
Replies in plain language with citations and the effective date of the policy version used.
- 04
Escalate
Hands sensitive or low-confidence cases to an HR partner with a clean summary of the conversation.
Capabilities
What HR actually does
Built for People ops, managers, and every employee with a question.
- 01
Guided onboarding
Walks new hires through day-one to day-thirty checklists with progress memory.
- 02
Cited policy answers
Answers leave, benefits, and process questions with source references and version dates.
- 03
Ticket deflection
Resolves common requests and routes the rest with full conversation context attached.
- 04
Human handoff
Escalates sensitive topics to HR instead of improvising policy it cannot verify.
Architecture
How the system thinks
Data flows left to right: sources feed the orchestrator, specialists reason in parallel, tools execute, and one artifact comes out.
Sources
- DataPolicy library
- DataHRIS
- DataOnboarding playbooks
Orchestration
- OrchestratorHR orchestrator
- LLMReasoning LLM
Specialists
- AgentOnboarding agent
- AgentPolicy agent
Tools & output
- ToolVector search
- ToolTicketing
- OutputCited answer
- Data
- Orchestrator
- LLM
- Agent
- Tool
- Output
In action
A real exchange
An abbreviated conversation, close to what the agent actually returns.
HR Buddy Session active
- You
How much parental leave do I get, and do I have to take it all at once?
- HR Buddy
You have 16 weeks at full pay, and it can be split into up to three blocks within the first 18 months. Source: Parental Leave Policy §3.2, updated March 2026.
- You
Can I start next month?
- HR Buddy
The policy asks for 8 weeks of notice, so next month would need an exception. I have flagged it to your HR partner with your dates so they can confirm.
Design decisions
Trade-offs worth naming
The constraints that shaped the system — and what was deliberately left out.
- 01
Never improvise policy
When retrieval confidence is low the agent hands off instead of guessing. A wrong HR answer costs far more than a slow one.
- 02
Citations by default
Every answer carries the policy section and version date, so employees and HR read the same source of truth.
- 03
Memory scoped to the employee
Conversation context never crosses employees, and sensitive attributes are excluded from retrieval.
Honest limits
Where it stops, where it goes
What the agent does not handle today, and what comes next.
Current limitations
- Only as current as the policy library — stale documents produce stale answers.
- Country-specific edge cases still need a human HR partner.
- No write access to payroll or HRIS records.
On the roadmap
- Multi-language answers with locale-aware policy resolution
- Proactive onboarding nudges instead of pull-only Q&A
- Manager-facing team summaries
Stack
Technology
Nothing added without a concrete requirement behind it.
NuxtLangGraphOpenAINotionpgvector