North Dakota Legislative Council | Llama case studies
LLAMA 3.2 1B INSTRUCT
North Dakota Legislative Council
<6 minutes
to summarize a draft bill
<5 seconds
to summarize a smaller bill
15%–25%
estimated time saved per bill summary
25 hours
estimated legal team hours saved per legislative session
100% on-premises
solution runs on secure, local hardware
*All results are self-reported and not identifiably repeatable. Generally expected individual results will differ.
CASE STUDY
How North Dakota’s legislative IT team developed a new, secure tool, built on Llama, to enhance lawmakers’ legislative process and to benefit North Dakotans
The North Dakota Legislative Council piloted an AI solution using Llama to review and summarize 1,089 draft bills, subsequent revisions and final legislation during the state’s 2025 legislative session. For the 2027 legislative session, this technology will free up valuable time for the council’s nine-member legal team to focus on the complexities of lawmaking and shape more impactful legislation.
At a glance
Industry
Government
Use case
Summarizing documents with sensitive information
Goal
Automate time-intensive review and summarization of in-flight legislation
Llama version
Llama 3.2 1B Instruct
Deployment
100% on-premises, running Ollama on secure, local hardware
Organization size
45 permanent legislative staff
THE CHALLENGE
A lean legal team was working overtime to summarize bills
The North Dakota Legislature convenes every two years for 80 days, which means they have a lot to do in very little time. Despite the time constraints, the legislature is busy, passing an average of 515 bills per session. In the 2025 session, 1,089 bills were introduced, each going through multiple revisions.The North Dakota Legislative Council (NDLC) — a small team of lawyers, accountants and IT professionals — serves the North Dakota Legislative Branch by drafting bills and amendments so that legislators fully understand proposed laws and citizens can participate in the legislative process. A key task of the NDLC is summarizing each proposed bill and publishing a comprehensive report — a task that requires an estimated 1,380 hours of expert labor, all performed immediately following an 80-day session with only 640 working hours. All of this must be done confidentially through the drafting process, which means the team had to build this capability in a secure, private framework.
THE SOLUTION
Llama-enabled AI solution automates bulk bill summarization
The NLDC IT team used Llama 3.2 1B Instruct to enable a secure on-premises summarization of the 2025 legislative session documents. Their application uses a retrieval augmented generation (RAG) workflow to ingest legislative drafts, revisions and final bills, plus prompts to generate structured queries for the model. For each bill, the solution generates three summaries for the legal team to review.
THE OUTCOME
Freeing critical time for North Dakota’s legal team
In the marathon environment of North Dakota’s biennial 80-day legislative sessions, every minute counts for the NDLC team of attorneys, accountants and IT experts. By automating review and summarization of bills, the NDLC IT team has lifted a major burden from the legal team that will allow them in the 2027 legislative session to dedicate more time to the complex, critical thinking required to craft effective legislation.
<6 minutes to summarize a draft bill
<5 seconds to summarize a smaller bill
CONCLUSION
Next steps: expanding automation across the legislative process
The NDLC aims to build on its success with bill summarization by expanding its Llama innovations to save time and address challenges across a variety of use cases. Future applications include developing a chat tool trained on NDLC’s website content, summarizing meeting videos to streamline minute-taking, and populating legislative dashboards with bill summaries, all with the goal of improving the legislative process to benefit North Dakotans.
Shopify | Llama case studiesShopify uses Llama to generate product pages, localize content, and automate support, helping developers scale workflows and save time.Read more
ConsumerTech
Scribd, INC | Llama case studiesDelivering faster, cheaper and more accurate results with a Llama-powered AI content discovery assistant.Read more
Tech
Exati | Llama case studiesTransforming support for smart city platform clientsRead more
The NDLC IT team wanted to give the legal team a head start on summarizing proposed bills and reduce its overall workload by using AI to review pending legislation and draft summaries.
Much of the NDLC legal team’s work happens before bills are made public, which made security and data control primary concerns. To meet stringent security requirements, the solution would have to run on local hardware, ruling out cloud-based AI services and API calls.
Proposed changes to the 65 titles in North Dakota’s Century code are assigned to nine legal staff members who work overtime throughout the legislative season. We hoped AI could take hours off their plates by reviewing proposed legislation and generating initial summaries.
Cody Malloy
Information Technology Services Director, North Dakota Legislative Council
Our team is excited. They see a tool that can help them save a lot of time, a tool they want to help improve and expand. The project has increased the collaboration between the legal division and information technology division — momentum we hope to keep up as we develop new capabilities.
Cody Malloy
Information Technology Services Director, North Dakota Legislative Council
The North Dakota Legislative Council uses Llama to make it faster and easier to summarize proposed changes to the 65 titles in North Dakota’s Century code.
Open source tools unlocked flexibility, control and cost savings
Llama and other open source resources like Ollama, Unsloth, and Hugging Face were essential for meeting the project’s budgetary, security and performance requirements. Equally important, the open source model of collaborative development and transparency matched the NLDC team’s approach to delivering legislative information to the people of North Dakota and their working relationship with other government agencies.
The portability of Llama plus Ollama’s simplified approach to running large language models (LLMs) on local hardware allowed the team to use local hardware for every step of development and deployment, ensuring data sovereignty. By using Llama 3.2 1B Instruct, the team reduced the solution’s memory and compute footprints, which allowed it to run on less expensive hardware.
Open source is very important to our IT team. Cost saving, collaborative development and transparency are major benefits. Having access to source code and being able to adapt weights with Llama and the help of resources like Hugging Face, Unsloth and Ollama made this project possible.
Cody Malloy
Information Technology Services Director, North Dakota Legislative Council
Fine-tuning Llama and deploying locally for secure and accurate summaries
Off-the-shelf Llama 3.2 1B Instruct outperformed other open-source models in initial summarization testing, but to handle the complexity of legal language and legislative summaries, the NDLC team fine-tuned the model for domain-specific accuracy.
The NDLC IT used Unsloth to fine-tune Llama. Unsloth is an open-source, fine-tuning tool that reduces training memory footprints and accelerates training time without sacrificing accuracy. For training, the NLDC team used previous legislative session bills and their expert-crafted summaries as their dataset.
After fine-tuning, the team used Ollama to deploy Llama and the summarization pipeline on local hardware. The prototype ingested draft legislation via a Retrieval-Augmented Generation (RAG) pipeline, enabling Llama to apply its general world knowledge to this specialized domain and successfully summarize relevant draft legislation. RAG is a valuable technique for loading just the relevant portions of the documents into the prompt, supporting the generation of high-quality, context-aware responses. This resulted in generating three summaries per bill providing the legal team options for creating the final summary.
As the 2025 legislative session came to a close, the system took less than 10 minutes to review 601 bills and resolutions and create three summaries for each, lifting a major burden from the legal team that will allow them in the 2027 legislative session to reclaim valuable hours and focus on the nuanced analysis and decision-making that shapes North Dakota’s laws.
The NDLC summarization service — powered by lightweight Llama 3.2 1B instruct — runs in an Ollama environment on secure, local hardware.
15%–25% estimated time savings per bill summary
25 hours estimated legal team hours saved per legislative session
100% on-premises — solution runs on secure, local hardware
*All results are self-reported and not identifiably repeatable. Generally expected individual results will differ.
In its most basic form, the legislative process is a legal document publishing service. Content authors — the legislators, legal team and analysts — produce drafts, summaries, reports and final laws. We’re using Llama to help all these authors process, generate and act on legislative information faster and more accurately.
Cody Malloy
Information Technology Services Director, North Dakota Legislative Council
Stay up-to-date
Our latest updates delivered to your inbox
Subscribe to our newsletter to keep up with the latest AI updates, releases and more.
Proposed changes to the 65 titles in North Dakota’s Century code are assigned to nine legal staff members who work overtime throughout the legislative season. We hoped AI could take hours off their plates by reviewing proposed legislation and generating initial summaries.
