This interview was conducted by Freddie Price (Senior Partnerships Manager, Apolitical), as part of a series for Apolitical’s 50 States, 50 Breakthroughs, created in collaboration with the National Academy of Public Administration and Humans of Public Service.

Maryland has a legislative dataset of almost 2,000 bills from the 2025 session and previous years. For small teams who must stay on top of shifting legislation, manually reviewing an entire dataset is not an option. Civil servants need a way to quickly identify bills relevant to their mission, rather than reading hundreds of documents.

Maryland’s Innovation Team, based in Governor Moore’s Office, built Legi-Assist to solve that problem. The tool solves this by downloading Maryland General Assembly bills, converting them into machine-readable formats, applying amendments accurately, and using large language models to answer policy-relevant questions about prior, current, and proposed laws. It helps public servants spend less time searching through dense documents and more time doing the work that matters.

In conversation with Apolitical, Francesca Ioffreda, Chief Innovation Officer for the State of Maryland, reflected on the motivations behind the project, how the team validated their approach, what they have learned, and what they are building next.

Tell me about the Maryland State Innovation Team and the types of projects you work on.

We’re a relatively new team within the state. I joined about a year and a half ago to build out this innovation team within Governor Moore’s office, formed through a partnership with Bloomberg Philanthropies. Our role is to create cross-cutting solutions that address Maryland’s most complex and entrenched challenges, starting with two of Governor Moore’s priorities: reducing childhood poverty and improving economic mobility.

You can think of the Innovation Team as a solution incubator within government. We’re a cross-functional team of eight, bringing together expertise in data science, civic design, and community engagement. That means we combine rigorous data analysis with direct community feedback to make sure our solutions are grounded in the real needs of Marylanders.

One of our most valuable roles has been acting as a bridge between different state agencies, and being in the Governor’s Office really helps with that. All of our efforts are executed in partnership with at least one, and often more than one, agency. At our core, we’re trying to fundamentally change the way government works: making sure that spending decisions, policy choices, and programmatic efforts are community-informed and guided by data, with a test-and-learn approach built in.

What problem were you trying to solve with Legi-Assist, and what made it urgent?

This project really began as a practical necessity. As a new team, we needed a way to quickly and reliably understand the policy landscape, both past and present, particularly as it related to child poverty. A broad scan of what had already been tried was essential. Understanding that historical context meant we could focus our efforts on interventions most likely to make a real difference for Marylanders.

The true urgency emerged during the legislative session. When you’re tracking almost 2,000 introduced bills and relying on manual review, it simply doesn’t scale. That’s what led us to create Legi-Assist: an automated pipeline that downloads legislation, cleans the text, and uses AI to analyze it.

What started as a helpful research aid quickly became a critical operational tool that allowed us to stay current, stay accurate, and keep pace with the legislative cycle without overtaxing our small team. We also realized, as we were developing it, that we weren’t alone. There are hundreds of dedicated civil servants across state government facing the exact same challenge. We saw an opportunity to help modernize their workflows alongside our own.

Before Legi-Assist, what did the workflow look like for analysts and policy staff, and where were the biggest time sinks?

There was a significant capacity divide. Large agencies might have dedicated legislative teams, while smaller offices have to fit manual bill review into already full workloads. That was a real constraint.

But the biggest risk wasn’t just inefficiency, it was incomplete analysis. Staff often had to rely on quick keyword searches or short bill synopses just to manage the volume. The problem is that a synopsis is a summary, and it can easily omit a specific agency mandate or critical detail. If an agency doesn’t know it’s named in the full text of a bill, it misses the window to submit testimony and shape the policy. Legi-Assist was built to prevent those silent misses and ensure more comprehensive visibility across the board.

What did the Innovation Team build, and what are the major components that make Legi-Assist useful day-to-day?

Foundationally, Legi-Assist is an automated pipeline that downloads Maryland General Assembly bills and processes them from complex PDFs into clean, machine-readable text. From there, it uses AI to analyze text, helping us to answer policy-relevant questions about existing and proposed laws.

One of the major breakthroughs is how Legi-Assist handles formatting that standard AI ignores. In Maryland, when a law is amended, the old text is shown with a strikethrough. Standard AI models read those crossed-out words as active law. We specifically engineered Legi-Assist to visually recognize and exclude those strikethroughs, so we’re not making policy decisions based on repealed language.

We also developed what we call a relevance engine. This allows the tool to recognize that a bill about school lunch subsidies is highly relevant to child poverty, even if those exact words don’t appear in the bill. That moves us from basic keyword searches to genuine contextual analysis.

The relevance engine works by building detailed profiles for every Maryland state agency, department, and team, capturing their day-to-day responsibilities and the populations they serve. When a new bill is processed, the system feeds the updated text into the AI model. Rather than hunting for matching words, it can identify why a department is affected and explain it clearly.

How did you design and validate your approach, and how did you ensure Legi-Assist was both accurate and reliable?

We built Legi-Assist as a prototype to solve a problem our own team was facing. Accuracy was something we were acutely aware of from the start.

To validate the tool, we tested it against three years of Maryland legislation: over 1,800 bills. We benchmarked performance using complex bills our team knew in detail, like the Access to Banking Act and the ENOUGH Act. We knew the tool was reliable when it consistently interpreted stricken text correctly and could distinguish between a bill that merely mentioned an issue and one that mandated specific action.

Once we validated the accuracy in-house, we presented Legi-Assist at Maryland’s AI community of practice, where different agencies come together to share innovation. There was a lot of interest, and that signaled to us that the tool had potential well beyond our team’s immediate needs. We were then able to validate that manual tracking and review of legislation was a shared pain point and iterate on the tool to make it as useful as possible for agencies across the state.

Building in-house rather than purchasing a vendor product was a deliberate choice. It minimized costs, allowed us to move faster in response to feedback, and meant we maintained full ownership of the data processing.

What results have you seen since launch, and what has worked well?

Legi-Assist now scans all newly proposed bills every night and delivers a prioritized, relevant list to more than 120 state agencies and teams every morning. During the legislative session, that alone has made a real difference.

But some of the most exciting results have been unexpected. We recently needed to help estimate the cost of a complex new data system. We used Legi-Assist to identify previous legislation that had established similar systems in the state, then analyzed the official fiscal notes on those past bills to create a data-backed historical baseline for the financial estimate. What would have been a rough guess became a research-evidenced projection.

Collaboration has also been central to how the tool has evolved. We’ve continued to learn from agency leaders that have spent years in state government, understanding what matters most in a legislative tool and where the old process fell short. That feedback loop is ongoing. A core philosophy of our team is that you have to design a prototype to get the right feedback, and then you have to continue to evolve alongside the people you’re building for.

Why was it important to make Legi-Assist open source, and what advice would you give to other governments looking to replicate this?

We made the code open source on GitHub because the challenge of managing overwhelming legislative volume isn’t unique to Maryland. It’s shared by governments across the country. We wanted to give other states the foundational infrastructure so they don’t have to build from scratch. It also advances transparency by giving residents a tool to track legislation that affects their communities.

For other government leaders, I’d offer two pieces of advice. First, start with a real problem you’re actually facing. Find a tedious, high-friction administrative task and solve it well. When you do, you unlock capacity you didn’t know was there. That’s often where the best ideas come from.

Second, invest in your internal workforce. I’m humbled every day by the talent of the public servants I work with. Give that talent the time and permission to build solutions, and then enable those solutions to scale. That doesn’t always require expensive outside contracts.

What is the Innovation Team working on next, and how are you leveraging new technology across your projects?

Last December, our team launched the Maryland Community Business Compass: a first-of-its-kind data platform designed to close critical gaps in essential services like childcare and fresh food in high-need communities. The Compass equips local entrepreneurs and community investors with data on where market opportunities exist and where gaps are greatest. It also provides a searchable database of over 250 state and local funding opportunities, alongside business support resources covering planning, permitting, and licensing.

One of the most exciting features is how we used AI to elevate community voice. There had been so much work done in communities across the state through local planning processes, and we wanted to honor it. Most of it lived in PDFs that weren’t regularly referred to. Using AI, we synthesized 200 community plans from across the state: nearly 20,000 pages of text, pulling out specific quotes and insights around resident-defined needs in childcare and fresh food. That provided additional validation and helped ensure our data reflected what communities themselves had identified as priorities.

We paired the launch with $10 million in new state investments from Maryland’s Department of Commerce and Department of Housing and Community Development for childcare and fresh food. Two local residents have already used it to identify childcare deserts in Howard County, analyzing local income data, costs, and existing childcare centers to find possible locations for a new center. They told us the tool gave them a running start and saved them from paying for expensive market research.

By putting this kind of data directly in the hands of communities, we’re democratizing access to information, and with it, access to opportunity. We’re now in the discovery and user research phase for the next version of the Compass: meeting with entrepreneurs, community partners, and state agencies to understand how they’re using the tool and shape what comes next.