IR Insights

IR Insights is an experimental, zero-cost proof-of-concept exploring how retrieval-augmented AI tools can support enterprise bargaining and workplace relations research in the Australian higher education sector.

The project was developed to address a familiar problem in enterprise bargaining: critical information exists across legislation, enterprise agreements, Fair Work Commission materials, and other public documents, but is often fragmented, difficult to synthesise, and time-consuming to navigate.

At its core, the project is a topic-specific, retrieval-augmented AI knowledge system for enterprise bargaining and industrial relations matters, using Google’s NotebookLM as the retrieval and grounding layer, and a curated front-end for simple navigation.

PURPOSE

IR Insights is designed as a research and exploration aid for employer-representatives and practitioners in the Australian higher education sector. It is intended to:

• support understanding and comparison of complex material

• assist with orientation to unfamiliar topics or processes

• help identify patterns, themes, and points of difference across documents

It is not intended to: provide legal, industrial, or professional advice; replace primary source analysis; or determine bargaining positions or outcomes. All outputs are experimental and should be treated as informational and should be validated against original sources.

AUDIENCE

While developed as a set of experimental research tools, IR Insights is intended to support a range of professional users involved in enterprise bargaining and workplace relations within the higher education sector. This includes experienced workplace relations practitioners undertaking detailed agreement comparison and analysis, senior leaders and executives participating in bargaining for the first time, and generalist HR professionals seeking to understand legal obligations or unfamiliar processes.

A key design objective of IR Insights is to demonstrate what can be achieved using entirely free, publicly available AI and related tools, without bespoke development, licensing costs, or proprietary datasets. This makes the project intentionally lightweight, replicable, and accessible, but also introduces important constraints.

In particular:

• all source material must be manually selected and maintained

• updates to legislation, agreements, or guidance require active curation

• outputs should always be understood as reflecting the point in time of the underlying sources

RETRIEVAL-AUGMENTED AI

Enterprise bargaining and workplace relations research presents a specific technical challenge as the relevant material is large, dense, and highly interdependent. Research may require consideration of dozens of enterprise agreements, awards and legislation running to hundreds, if not, thousands of pages. Together, this represents an extremely large context window, far beyond what traditional search tools or document viewers handle effectively.

NotebookLM is well suited to this challenge because it can:

• ingest and reason across very large document sets simultaneously, beyond the capabilities and context limits of other general chat-based AI such as Claude, ChatGPT and Gemini.

• interrogate content holistically, rather than page-by-page

• identify conceptual similarity, not just exact wording

This allows the system to recognise and collate provisions that use different terminology to describe similar concepts — a critical capability when comparing enterprise agreements drafted in different styles and contexts.

SOURCE MATERIALS

Sources included for each research notebook are drawn from only publicly available and authoritative materials, including:

• legislation and regulations

• Fair Work Commission publications and bench books

• enterprise agreements and awards

• publicly available union logs of claims

• selected explanatory and guidance materials

The inclusion of a document reflects its relevance to the topic area, not an endorsement of any position or interpretation. The project does not aim to be exhaustive; instead, it focuses on representative and practically useful material to support comparative and exploratory research.

CAPABILITIES BEYOND CONVERSATIONAL CHAT

Although conversational querying is the most visible feature, NotebookLM also supports a range of structured research outputs that are particularly useful in enterprise bargaining contexts. These include:

• comparative tables of clauses or provisions

• thematic summaries across multiple documents

• briefing-style slide decks

• visual summaries and infographics

• audio or “podcast-style” explanations of complex topics

These outputs are intended to support briefing, education, and exploration — not to replace primary source review or professional judgement.

General AI tools such as ChatGPT, Gemini, or Copilot are powerful and with the right guidance may produce similar results. However, they are not optimised for context-constrained, document-specific research.

The choice of Notebook LM differs in several important ways:

• Closed source sets - Each notebook operates over a defined, curated body of documents. The AI does not draw on general internet knowledge or unrelated material.

• Source reliability and traceability - Responses are anchored to specific documents, with clear references to the underlying source material. This allows users to verify outputs and build confidence in the results.

• Large, topic-specific context windows - Rather than querying one document at a time, the system can reason across dozens of agreements, awards, or guidance documents simultaneously.

• Conceptual rather than keyword-only search - Compared to traditional research methods, the AI can identify related concepts even where different language is used, reducing the risk of missing relevant material due to drafting variation.

For research, comparison, and orientation tasks, this approach is often more reliable and interpretable than general-purpose AI querying that may more readily drift; rely on unverified or unreliable source material.