Legi-Assist

What’s the Problem?

Every legislative session, the Maryland General Assembly (MGA) introduces around 2,000 bills. State agencies must track all incoming legislation to learn if a new law changes their mandates, alters their funding, or assigns new responsibilities. Catching these updates early allows agencies to submit testimony and prepare their staff. An agency can’t easily adapt if they miss a bill that affects them, which in turn hurts the citizens who need their services. 

Historically, agency teams have relied on manual keyword searches and short official synopses. However, these methods are not easily scaled across large volumes of legislation. This presents three core challenges for state agencies: 

• Capacity Overload: Manually reading 2,000 complex bills is an impossible task, especially for smaller agency teams with limited staff. When teams are overwhelmed, critical legislation slips through the cracks unnoticed. If an agency doesn't even know a bill exists, they can't respond to it or plan for its impact. 
• Inefficient Vetting Methods: Basic keyword searches only flag bills that contain exact phrase matches, which makes it easy to miss highly relevant bills that use slightly different phrases. On the flip side, official bill synopses are too brief; they cover general themes but often leave out specific, fine-print clauses hidden deep in the full text. 
• Visual Formatting: MGA legislation uses visual strikethroughs in PDFs to show when a law will be deleted or repealed. Standard AI ignores this formatting during text extraction, meaning it mistakenly processes dead, repealed laws as active, incoming mandates. This leads to teams doing analysis on null policy, which results in more time spent to correct their mistake. 

In sum, state agencies miss vital legislation when they are trapped between staffing limits and flawed vetting tools. This impacts their ability to address new mandates and offer effective constituent services. 

The Innovation Team's Solution

Legi-Assist, built by the Governor’s Innovation Team, is an open-source Python toolkit that collects and analyzes State legislation. The tool automatically downloads PDFs of newly introduced bills and converts them into clean, machine-readable formats. From there, it uses large language models to analyze the text's actual substance—turning dense legal writing into clear and actionable language. 

Legi-Assist automates the heavy lifting of legislative review by providing bill summaries in plain English, sponsorship and status tracking, policy tagging, stakeholder analyses, and fiscal impact reports. The toolkit uses a custom relevance engine to automatically match bills to their respective agencies, removing the need for keyword searches and official synopses. 

Legi-Assist's built-in visual recognition catches PDF strikethroughs, ensuring that the tool only processes active legislation. A user-friendly web interface ensures that non-technical civil servants, agency leaders, and policy analysts can easily access these features. 

How Does Legi-Assist Work?

Legi-Assist relies on three core pillars to analyze legislation: Automated data sync, agency relevance scoring, and visual recognition. 

• Automated Data Sync: Every night, Legi-Assist connects directly to the MGA database to download and index newly adopted amendments, the latest bills, and updated fiscal notes. This automated process ensures that agencies always have the most up-to-date legislative data each morning.
• Agency Relevance Scoring: Matching legislation to the right department goes far beyond basic keyword matching. The Governor's Innovation team built a written profile of over 120 State agencies directly into Legi-Assist's system. This profile details specific mandates, daily work, and the populations that each agency serves. The tool cross-references incoming bills against these detailed profiles to quickly score and route legislation based on best fit. 
• Visual Recognition: Legi-Assist's code is trained to accommodate the layout of Maryland's legislative PDFs, meaning it accurately identifies formatting like strikethroughs and excludes them from the analysis. This ensures deleted or repealed text is ignored and that relevance scoring is performed exclusively on active language. 

While the Innovation Team initially built Legi-Assist using the Gemini 3.0 Flash AI model, the toolkit’s architecture includes flexible, built-in utilities that allow Legi-Assist to support OpenAI, Google Gemini, or other open-source models run locally through Ollama.

Legi-Assist features an idempotent pipeline. (e.g., running one input multiple times will still yield the same output). Instead of processing all existing pieces of legislation from scratch each night, the tool checks the database for only new or modified bills. This saves both time and application programming interface (API) costs. 

How is Legi-Assist Compliant with Maryland's AI Governance Policies?

Legi-Assist was designed, built, and tested by Maryland state employees as opposed to a third party vendor. This internal development strategy supports the State’s mandates for responsible, secure, and trustworthy AI use detailed in Gov. Wes Moore’s Executive Order on AI. 

By leveraging generative AI coding assistants under human oversight, the Innovation Team followed the State’s “Human-in-the-Loop” protocol outlined in Maryland’s responsible AI implementation guidance. This framework treats AI tools as assistants rather than autonomous decision-makers, meaning state developers manually reviewed, analyzed, and approved Legi-Assist’s code to maintain accountability for the software's security.

Developing the tool in-house allowed Maryland's Innovation team to control its costs, speed, and level of control. Because the tool was built without a traditional procurement process, the State retains full ownership over how it processes legislative data. This also allows the Innovation Team to quickly deploy updates or revise the code when agencies have requests. This direct pipeline between user feedback and software improvement allows the State to serve more people in less time.

Key Impacts

Currently, Legi-Assist’s external version is transitioning out of pilot testing and supports over 200 active users across governments nationwide. This user base continues to grow as the Innovation team prepares to conduct more robust security testing for the system's user-friendly frontend. 

For the State of Maryland, the tool has had major operational impacts:

• Ranked Daily Assessment: Every morning during Maryland's legislative session, more than 120 state agencies receive a tailored slate of relevant bills generated from the overnight data sync. This customized docket ranks bills in order of relevance, allowing agencies to see at a glance how incoming legislation affects their daily operations. 
• Fewer Missed Bills: Because the custom relevance engine cross-references the full text of all incoming bills against its agency profiles, it successfully identifies agency mentions or specific duties that keywords and synopses often miss. This automated matching significantly reduces the likelihood that critical, fine-print policy changes slip through the cracks unnoticed. 
• Equalizing Agency Bandwidth: Smaller agencies can now operate with the same efficiency as larger departments. By handling the heavy lifting of text processing, Legi-Assist allows smaller teams to track policy shifts as easily as larger departments. This empowers them to serve more Marylanders in less time, with more accurate information.

The Innovation team wanted to give other states a foundational infrastructure so they don't have to build their own tracking systems from scratch. Because of this, Legi-Assist is fully open-sourced on GitHub under a permissive MIT license to make sharing the toolkit as frictionless as possible. The repository saw over 160 clones in its first two weeks online. 

Constraints and Limitations

Legi-Assist is built specifically for civil servants, agency leaders, and policy analysts who track state legislation and its direct policy impacts. The tool is currently in an internal pilot phase and is not yet available to the general public. This ensures that state teams can thoroughly evaluate the tool’s stability before releasing it for constituents to use.

From a technical standpoint, the toolkit’s infrastructure requirements limit how it can be deployed. Operating the software requires Python 3.10 or higher, alongside the precise dependencies outlined in the project's setup files - TXT. Because the tool relies on large language models to analyze text, agencies must also provide their own API keys for commercial providers like OpenAI or Google Gemini.

Legi-Assist can connect to cloud-based artificial intelligence providers as well as supporting open-source models run locally on an agency's own servers using Gemini, OpenAI, or Ollama (a local engine that allows agencies to host open-source models on their own servers). While models that fall outside of these specific parameters are not compatible with the toolkit's pre-configured options, users may clone the repository and create a connector to another preferred model provider if they choose.

Demo of Legi-Assist

The following screenshots display Legi-Assist's user interface and its various functions using Maryland's Artificial Governance Act of 2024:

Legi-Assist landing page:

Searching and filtering for the AI Governance Act of 2024:

Summary, analysis, and tags for the AI Governance Act of 2024:

What's Next?

Legi-Assist was recently selected as Maryland’s featured innovation in Apolitical’s "50 States, 50 Breakthroughs" showcase. Evaluated by a panel of expert practitioners, the tool was recognized for its real-world public sector impact and its potential to be scaled by governments nationwide.

The software itself is currently moving through two distinct stages of deployment. As of August 2026, its user-friendly interface is transitioning out of its internal pilot testing phase to prepare for broader, active use across state agencies. At the same time, the project's underlying backend code is already fully functional, allowing State developers across the U.S. to build upon and deploy the tool.

Looking ahead, the Innovation Team is considering how they might use Legi-Assist to predict system costs and fiscal impacts. The team discovered that Legi-Assist can help estimate the budget for a complex new data network by analyzing prior laws that built similar systems, then cross-referencing them with previous fiscal notes. This allows teams to build research-backed cost projections that are rooted in the State's own records, rather than relying on rough estimates.

Maryland state employees and contractors who want to explore this repository or learn how to leverage similar architectures can contact Governor’s Innovation Team’s Chief Innovation Officer, Francesca Ioffreda, or otherwise direct inquiries to the Innovation Team’s general inbox.

Maryland's Department of Information Technology (DoIT) also hosts free office hours every Monday. DoIT's AI enablement experts  can help answer questions about Maryland’s responsible AI use policy and to assist in rolling out secure AI utilities across State agencies.