Quick Recap

Yesterday, our guest Karen Suhaka, founder of BillTrack50, presented with us on using AI to analyze legislative data, highlighting how her platform provides bill summaries, tracks legislation, and aids public understanding. She discussed the complexities of the U.S. legislative process, the technical and ethical considerations in building policy-focused AI tools, and advice for entrepreneurs creating civic tech solutions.

Karen has generously shared both her BillTrack50 API and her slides for the presentation with us. To access the API (until August 1, 2025), you must reach out to Karen@LegiNation.com as a competitor in AI4Legislation. Thank you, Karen!

Featured: Karen Suhaka, serial entrepreneur and founder of BillTrack50

Introduction

SVCAF invited Karen Suhaka, founder of BillTrack50, to present in our second webinar in AI4Legislation. Karen is a serial entrepreneur with two previous successful ventures, and founded BillTrack50 in 2011 to make legislative data more accessible. Since her prior background was in physics and mathematics rather than legislation, the journey this time involved learning deeply about a new field.

Legislative Process Refresher (from a data perspective)

Karen started off by explaining how a bill becomes a law:

1. An idea from a citizen, lawmaker, or interest group is brought to a legislator for sponsorship.
2. The legislator submits the idea to a bill drafter, who formats the proposal into a compliant legislative draft. Some states allow access to draft bills but most do not, which hinders early data collection.
3. Once finalized, the bill is introduced to the legislature, read at the House or Senate where it originated from, and requires the support of a committee to proceed to floor voting.
4. If its originating chamber’s floor vote passes the bill, the process must be repeated at the other chamber. Any amendments must be agreed upon by both chambers until a final version is approved. The floor vote data are typically made available
5. The Governor or President can sign the bill into law, veto the bill, or let it sit until it becomes a law automatically or dies (pocket veto). If vetoed, the legislature may override the veto with a higher vote threshold. Veto and override data are typically publicized.

Across the 50 states, each legislature functions a little differently, and the data (such as votes, bill texts, and committee summaries) that are available generally vary in format as well. These inconsistencies add up, resulting in a highly complex process to track legislative bill processing across the nation.

Building Legislative Tools with AI

Key areas that developers should consider include defining the painpoint, knowing your audience, addressing data challenges, choosing AI models, and considering ethical standards. Karen emphasized that a majority of the project occurs before any programming happens, because the most important aspect of a successful business venture is making sure people have a need for your solution, and you have a way to market your solution to that target audience. Legislative AI entrepreneurs must also consider variances in data formats (HTML vs. PDF, structured vs. unstructured, costs and reliability), test different AI models or vector databases, and honor the ethics of AI usage (such as privacy, partisanship, bias, human error, misinformation).

Demo: How BillTrack50 Uses AI

Next, Karen demonstrated how AI generates plain-language summaries for bills. Vectorization can be used to show similar bills across states. BillTrack50 also includes user tools like flagging inaccurate summaries, which helps improve the product. Karen acknowledged that occassional weird AI outputs may appear, and explained that they are filtered or corrected in different ways.

Project Ideas for AI4Legislation Participants

Karen raised a few project ideas for competing in AI4Legislation:

• Predict the likelihood of bill passage
• Analyze intent or sentiment of a bill (e.g. anti vs. pro gun control)
• Explain amendments and their impacts on existing laws
• Track model bills that spread across multiple states
• Compare how legislators vote vs. constituents’ actual values

Tips on Entrepreneurship

Karen had three main tips for successful entrepreneurship:

1. Build a Minimum Viable Product (MVP) – what is the smallest feature set that someone will pay for? You want to get your first non-friend customer quickly, and sometimes that means the product might not be completely “finished” yet.
2. Form a team that fits your company’s culture, not just filling in gaps in skillsets.
3. Differentiate between “dumb money” and “smart money”
   • “dumb money” – your family and friends might help fund your project, but that’s usually a one-time thing with no buy-in or further benefits attached.
   • “smart money” – find investors with industry knowledge and not just money! They can connect you to other resources that can help you grow your business.