Case Study 11
AI Assistant / Slack

AI HR Assistant:
Inside Slack.

An HR assistant that lives in the company's Slack. It answers policy questions from the company's own documents and checks time-off requests against the rules before a manager is asked to approve them.

Role

AI Product Engineer

Stack

TypeScript, Node.js, Slack API, LLM

Channel

Slack

Scope

Policy Q&A, Time-Off Checks

AI HR assistant in a Slack thread pre-checking a time-off request for balance, notice period, team overlap and blackout dates, with a cited policy answer

Fig 0.1 – A time-off request pre-checked in Slack, with the policy source on every answer.

01 // The Problem

HR as a
Help Desk.

The HR team spent a large part of every week answering the same questions in direct messages. How many days do I have left. Can I carry them over. What is the policy for working abroad. Time-off requests had a second problem. They reached managers without context, so conflicts with the team calendar, a short notice period or a missing balance were found after approval, or by HR cleaning up afterwards.

02 // What I Built

Answers From
the Handbook.

The assistant sits where people already ask: in Slack. It answers from the company's own HR documents and shows the policy section each answer comes from, so people can check it instead of trusting it. When the documents don't cover a question, it says so and hands the thread to a person in HR rather than inventing an answer.

03 // The Hard Part

The Bot Checks.
Humans Approve.

Time off is where an AI assistant can do real damage, so the split of responsibilities was the main design decision. The rules themselves, such as remaining balance, notice period, team overlap and blackout dates, run as ordinary code against the company's data, because they need to be right every time. The language model only reads the request and explains the result in plain words. The request then reaches the manager with the pre-check attached. The assistant never approves anything; the final decision stays with a person.

04 // Guardrails

Private by
Default.

HR data is personal, so access follows the person asking. An employee can ask about their own balance, not a colleague's, and the assistant only sees the documents it needs to answer. Conversations are logged so HR can review what the assistant said and improve the source documents where the answers were weak.

The Build.

What changed for HR and for managers, described by how the system works rather than by a marketing report.

Slack
Where Employees Ask
Cited
Policy Source on Every Answer
Pre-Checked
Every Time-Off Request
Human
Makes Every Final Approval
Next Project

WordPress to HubSpot Migration

→