BambooHR AI Chatbot — Policies & Benefits Assistant

Overview

Employees regularly need answers about policies, benefits and who to contact — information usually scattered across HR documents and directories. This project is an AI HR assistant that brings it together in a single chat interface on top of BambooHR.

How it works

An ingestion workflow pulls company documents from BambooHR, loads the PDFs, splits them with a recursive character text splitter and stores OpenAI embeddings in a Supabase vector store.

The chat workflow runs an AI agent with window buffer memory and two main tools. The first is a vector-store tool that retrieves relevant passages for policy and benefits questions. The second is an employee-lookup sub-workflow: a text classifier decides whether the user is asking about a specific person or about a department, and the workflow then queries the BambooHR employee directory accordingly. For department questions, it identifies the most senior person by walking the reporting structure, using an information extractor and LLM chain to interpret the results.

Structured and auto-fixing output parsers keep tool responses well-formed, and when the documents cannot answer a question, the assistant escalates by pointing the employee to the appropriate HR contact.