Each year, the National Labor Relations Board (NLRB) and the courts issue thousands of decisions, legal briefs, and memos that shape the labor rights of millions of Americans. But navigating this sea of legal information is both difficult and time-consuming, making it hard for workers and their advocates to understand their rights.

This inaccessibility has real consequences for workers.

For example, a worker at a nonprofit organization was in the middle of their union's first contract negotiations when management dropped a bombshell: they planned to eliminate a cost-of-living adjustment that had been standard practice for years. 

None of the union stewards realized that the employer's threat was illegal under federal labor law – until the worker stumbled upon a relevant case while reading NLRB Edge, a free, AI-powered newsletter that summarizes new labor law decisions. The worker then used this information to educate the bargaining team and demonstrate why filing an unfair labor practice charge was justified under that legal precedent.

What could have been a significant loss of benefits was avoided, but only because someone happened to find the right information at the right time.

This near-miss reveals a fundamental problem with how the NLRB makes labor law information available to the public and points toward a solution.

The Problem: Fragmented and Unusable Legal Information

The National Labor Relations Board is the independent federal agency that interprets, enforces, and administers the National Labor Relations Act, the law governing private-sector labor-management relations in the United States. 

The Board's regional offices supervise union representation elections, while the General Counsel investigates and prosecutes alleged unfair labor practices before administrative law judges (ALJs). The Board then hears these cases and issues rulings that create a body of legal precedent that workers, union organizers and advocates, lawyers, policy researchers, and policymakers all need to understand workers' rights under the law.

But accessing this information has long been unnecessarily difficult. The NLRB’s official website lists different types of documents – including ALJ decisions, Board decisions, regional election decisions, and others – as PDFs on separate webpages. While some of these pages are searchable, others aren’t, and there’s no way to search across all of these documents at once. Other relevant documents, such as federal court rulings that review NLRB decisions, are on separate websites altogether. 

These limitations create practical barriers to legal research: individuals must know which category of document they're looking for before they can begin searching, wait through painfully slow load times, and navigate multiple sites and pages to piece together a complete understanding of how labor law applies to any given situation.

Making Labor Law Knowledge Searchable

Enter Matt Bruenig, a labor lawyer who saw an opportunity to use newly available AI platforms and open-source programming tools to make this government legal information more accessible to those who need it. 

NLRB Research is a free online database that consolidates labor law information into a single searchable platform. The self-updating database compiles more than 116,000 documents, including NLRB decisions and relevant court cases. 

The database includes several features that make it easier to sift through this wealth of legal knowledge. In addition to searching for keywords in document titles and text, the database can be searched and filtered by attributes such as publication date or case number. Each document is accompanied by an AI-generated summary. The platform also includes a citation-linking feature that enables users to easily identify documents that cite a particular case. 

The nonprofit worker's story, shared in an email to Bruenig, illustrates how the database and the newsletter it powers help to make valuable labor law knowledge accessible to those who need it most. 

For labor lawyers, legal researchers, and policymakers, the ability to rapidly search across the complete set of NLRB decisions enables analysis that wasn't practical before, from tracking how specific doctrines evolve over time to identifying patterns in regional enforcement.

How it Works

To understand how the database can be used in practice, let’s look at an example. Say I were researching the legal basis for the growth in union membership among graduate students over the past decade.

I could begin by researching the 2016 Columbia University decision, a seminal case in which the Board held that graduate student workers at private colleges and universities were entitled to collective bargaining rights under the National Labor Relations Act.

Using the search tool and filters, I can quickly pull up the published Board decision, while reading the AI-generated summary helps me get up to speed on what the case is about. 

From there, I can search by the case number associated with the decision (Case 02–RC–143012):

That search pulls up other documents related to the case. For example, we can see that the case originated with a 2015 unpublished Board decision directing the Regional Director to hear a petition brought by the union, Graduate Workers of Columbia, that had previously been dismissed without a hearing. 

I can also click on the citation to find documents that cite the Columbia University decision. 

This search reveals several recent cases where regional offices applied the Columbia University standard – which established criteria for determining whether temporary employees share sufficient common interests to engage in collective bargaining – in their rulings on union petitions at the University of Pennsylvania, Brown University, and Loyola Marymount University.

This kind of research – tracing how a legal precedent emerged and evolved across multiple cases – would take a researcher hours using the NLRB's official website, if it were even possible at all. With NLRB Research, it takes minutes.

From Experiment to Essential Resource

The NLRB Research database originated as an experiment with large language models. 

"Initially, I wasn't trying to solve a problem, but rather, was trying to see if I could use LLMs to help me with legal research," says Bruenig, who describes himself as having a "hobbyist programming background."

His initial vision was to create an AI-powered chatbot that could answer questions about labor rights based on the Board’s past decisions. 

To gather all the relevant documents in one place, he wrote scripts using the programming language Python to scrape documents from the NLRB website and court rulings from the legal database Justicia. The scripts automatically downloaded this information in bulk, converted PDFs to plain text, and organized it into a structured, self-updating database.

By the time Bruenig built the initial version of the database, he realized the limits of what he could do with LLMs when working with 40 gigabytes of legal text. His original chatbot vision hit technical constraints around context windows – the amount of information that an AI model can process at one time – that even today's more advanced models struggle with.

But even without the AI features he'd envisioned, the database was massively more searchable and accessible than the NLRB's official site. It also included documents that even paid legal databases such as Westlaw and Lexis didn’t have. So rather than abandon the project, he decided to make it available as a free public resource.

The initial functional version took approximately six weeks to develop, with Bruenig working on the project in his spare time. He refined the initial version based on feedback from over 100 volunteer testers, including legal scholars, labor lawyers, union representatives, and rank-and-file workers, and launched the public resource in January 2025. 

The Tech Stack: Combining Open-Source Tools with AI

NLRB Research is powered by a combination of open-source tools – software which can be freely used and modified – and paid AI tools.

The project is hosted on a server running the open-source operating system Ubuntu. For the interface, Bruenig uses Datasette, free software that turns a SQLite data file into a searchable website. He continues to use Python scripts to automatically scrape new documents from the NLRB website and relevant court decisions from CourtListener.

To create the AI summaries, Bruenig uses the Gemini Flash 2.5 model via Google’s application programming interface (API). An API allows different software programs to communicate with each other – in this case, allowing Bruenig's programming scripts to automatically send documents to Gemini in bulk and receive summaries in return.

Bruenig chose Gemini because it "offered the lowest cost with a context window large enough for NLRB decisions." He created a customized prompt for each of the 14 document types. For example, published board decisions often contain both an ALJ's recommendation and the Board's response, so Bruenig's prompt instructs the model to distinguish between these two decisions and highlight any disagreements.

The database also powers NLRB Edge, Bruenig's Substack newsletter that tracks developments in labor law. Each morning, the database generates a list of newly added documents, which Bruenig runs through the Claude Sonnet 4.5 AI assistant to produce a summary.

He then uses Google's Nano Banana Pro model to generate an image illustrating the case and its outcome. Finally, he runs the list of cited cases through the database's citation search tool to automatically generate links to the relevant precedents. The entire process takes about 15-20 minutes each morning.

While NLRB Edge is free to read, the publication has approximately 250 paid subscribers who choose to support the project. 

NLRB Research costs approximately $600 per year to host and maintain, excluding Bruenig’s time. However, he says that the amount of money earned through paid NLRB Edge subscriptions more than covers the cost of running the project. 

Looking Ahead

Bruenig hasn't abandoned his original vision of an AI-powered labor law assistant. As language models and related tools like NotebookLM – which lets anyone create AI-powered knowledge bases from their own documents – continue to advance, his initial goal of building an AI that can analyze workplace situations and pull relevant case law may become feasible.

"I feel like we're getting closer to being able to do more of the right analysis," Bruenig says.

For now, the maintenance burden remains minimal, with Bruenig performing little work on the website other than troubleshooting occasional technical issues.

As AI tools become more powerful and open-source software continues to improve, projects such as NLRB Research point toward a future in which accessing government legal information doesn't require expensive subscriptions or technical expertise.

For government agencies tasked with serving the public, the question is not whether better is possible, but how to make it a reality.