For more than a decade, Shaun Brown has been building tech tools inside the House, first for Democratic leadership staff and then with House Digital Service. He shares responsibility for building innovative apps like DomeWatch and LegiDex.

Recently, he launched a new external product called Apogee to apply AI agents and algorithms to connect disparate streams of official legislative branch, federal election, and lobbyist disclosure data into more than 60 specific capabilities for advocates, nonprofits, policy wonks, and congressional staff.

Instead of dashboards or search commands featured in legacy legislative tracking and intelligence services to access public data in a gated environment, Apogee pairs with leading agentic AI platforms to allow users to ask questions that release agents to find information within official datasets. Asking “tell me everything about this bill” returns not only congressional action, but information on hearings, lobbying activity, news coverage, and advocacy groups’ positions. Agents can be set to watch topics as legislation advances or Congress takes up specific issues and provide users with updates automatically.

Apogee also generates knowledge graphs that can identify members’ political networks and influence, voting blocs and potential unusual voting alliances, interest group influence, and bill similarity, among other capabilities.

Brown sees AI’s role in dramatically reducing the difficulty in accessing and using public datasets to generate clear awareness of what’s happening in federal policy (he’s also building out state-level capacity). Applying existing AI models to official data from a variety of sources essentially levels the playing field for those working outside of well-resourced advocacy shops by producing reliable deliverables quickly and cheaply. Its cheapest plan starts at $29 a month, which fits neatly in member office tech budgets.

The value for congressional staff, Brown told me, is in the ability to generate the standard short reports, backgrounders, and one-pagers quickly and with information pulled directly from sources that can be verified by staff for accuracy. Its ability to track issues as well as members’ positions and legislative activity over time, meanwhile, can solve for some of the institutional memory challenges in member offices given high staff turnover.

Much of what Brown has done with Apogee could be built by Congress using its own data. House Clerk and Library of Congress staff discussed the potential of using semantic search to make Congress.gov more user-friendly in a recent CHA Modernization & Innovation Subcommittee hearing, so ideas are circulating within the legislative branch. For now, those on the outside continue to fill the gaps.

We’re going to share a project that’s a much larger scale than our previous entries this week to introduce Civic’s Revere, a new CRM going through the approval process with the House. Civic co-founder Jon Kokot recently explained to me that much of the product is designed around the pain points of current systems, which he heard about from his contacts serving as a naval congressional liaison. The major culprit in current CRMs was staff having to manually extract data from a tidal wave of inbound information to make it usable. He drew from examples of the recent explosion of AI-enabled fintech products as a comparative point for the congressional market, where the automation of data processing from a range of sources can free staff to focus on higher-level tasks, which in the congressional case is legislative efforts and casework. Jon gave a demonstration of Revere at last year’s Congressional Hackathon.

Revere operates closed AI models running locally to streamline processing constituent correspondence and casework management in a secure environment. Because it’s built on fresh distributed architecture, it can run multiple high-volume tasks at once without affecting performance. Models take inbound communications, apply office tags, batch them topically, and compile the constituent record. They also move casework out of a ticketing system, adding data and documents as a case heads through an agency’s process for centralized management. Kokot believes the system can cut down on casework delays ultimately if both congressional and agency staff interact through Revere.

The platform also allows member offices to run queries against the data in their CRM to support the trendspotting and review they routinely perform much more efficiently. Looking across the data, an AI knowledge manager assistant can perform sentiment analysis on issues constituents wrote in on, identify emerging policy concerns, and identify agency sticking points.

Modern communications technology has been a paradox for government, creating new inefficiencies in workflows when efficient communications methods create greater and greater scale. We’ve been hopeful that new innovations could break this paradox for some time, usually by layering on a specific fix to legacy systems. Civic is making more of a fresh start, which is long overdue and now much more feasible through AI.

We recently met a rising senior at Washington & Lee University named Sam Cook. As a Computer Science and Politics major, he observed the delicate dance between platforms and spreadsheets used by the congressional office in which he was interning to manage its scheduling as a potential project. He started on building a platform that could pull information from multiple sources into one scheduling dashboard as an independent study project last semester at W&L and returned to DC this summer to complete the work.

His HillDesk platform features a combination of automation and user review we think will save member offices significant time with necessary reassurance that information won’t be missed or garbled. It finds and flags emails containing meeting requests and event invitations through natural language processing and extracts the calendaring outputs side by side with the email, allowing users to double-check at the moment of processing. Cook has provided context based on a typical Hill office for the extraction system to improve its accuracy. The dashboard also pulls in the House floor schedule via DomeWatch.

In addition to being a promising tool to save congressional schedulers’ time and sanity, we’re intrigued by this project because it’s the first we’ve come across testing out the House Chief Administrative Officer’s unsolicited tech pitch form, which was announced at this spring’s Congressional Data Task Force meeting to provide an on-ramp for independent developers. If approved, it would be a great example of someone taking their experience from a congressional office and applying their know-how to make an incremental but meaningful improvement. That’s a sign of the kind of culture we’ve been championing in congressional modernization for some time. It’s even cooler that it’s coming from a college student who touched the institution for a short time but feels the motivation to make it better.

Reach out to Sam about the project at Samuel at hilldesk dot app.

This is your irregular update on tools, projects, and events concerning Legislative branch data and modernization. 

The Congressional Data Task Force met on June 11 for a series of presentations, demos, and public questions. Among the highlights: the Task Force unveiled a legislative branch data map; CRS demoed its text analysis program; the Secretary of the Senate announced app tracking vote alerts; and the CAO presented four projects that just received Modernization Initiative Funding. We at AGI showcased our efforts to track congressionally mandated reports. Read more here.

The House Administration Committee recently held two hearings: one on CRS & the future of AI-enabled policy analysis and the other on modernizing public access to legislative data. We wrote a history of CRS’s evolution, summarized the CRS hearing focused on its efforts to use AI, and analyzed the landscape of legislative branch data availability following the legislative data modernization hearing. Our written testimonies focused on reimagining CRS in the AI era and strengthening the Congressional Data Task Force.

We are continuing our makerspace blog post series, which focuses on people building congressional technology tools. Our most recent article covers Dave Zvenyach’s efforts on our behalf to use technology to verify congressionally mandated reports. Don’t miss our recaps of MyCapitol.ai, Capitol Labs, and Congress Press. If you are building a legislative-centric tool, we want to talk to you. Email congressionaldata@americalabs.org.

How are parliaments around the world using AI? I recently drew upon recent convenings that we co-hosted to discuss AI, Parliamentary Modernization, and the Legislative Plumbing of Democracy. 

The House Administration Committee published the most recent round of Modernization Reports in late June. They contain updates from support offices on various tech projects. They include updates on U.S. Legislative Markup, the Comparative Print Suite, Committee Votes and Tools, and Unique IDs for reports filed by registered lobbyists. We anticipate summarizing these reports in the First Branch Forecast newsletter and archiving that summary on the Congressional Data Coalition blog. 

The Modernization Subcommittee announced yesterday a series of exciting updates, including the launch of the House AI Innovators Pipeline to empower House staff to build custom applications for congressional operations; the securing of funds for House Rules digital infrastructure modernization— PRISM, to improve amendment intake, document generation, and statutory reporting; and the request for funds to develop “My Intelligent Assistant,” a customizable AI agent connected to House data sources to assist staff with their work. 

Our friends at Bussola Tech are continuing to host a series of online conversations about parliamentary digitization. Upcoming events include: Cybersecurity in the AI era and what makes digital transformation succeed in parliaments. A recent event of interest concerns the accuracy of speech-to-text technologies. Their news page is here.

Mark your calendars. 

The Congress.gov public forum is set for Sept. 24, 1-3 p.m. ET. The hybrid event, hosted by the Library of Congress, offers congressional and public stakeholders an opportunity to share suggestions for additional features for Congress.gov with the Congress.gov team and learn about changes to that service. It’s not yet possible to RSVP, but information is available here. We will likely host a happy hour afterward; details are TBD.

The Congressional Data Task Force is anticipated to hold its third quarterly meeting in September, but a date has not yet been announced on September 10 from 2-4 p.m. The fourth quarter meeting is scheduled for December 3. 

A Congressional Hackathon is also no longer anticipated for September, but a date has not been announced.

Information about those events will be published on the Task Force’s Innovation Hub webpage and the Congressional Data Coalition website.

If you have a project or announcement to share, or just want to talk about Congressional data, please let us know! Email us at congressionaldata@americalabs.org.

Congress routinely requires federal agencies to report back on their activities, but the legislative branch struggles with tracking and sharing those reports. After a decade long effort, in late 2022 Congress enacted the Access to Congressionally Mandated Reports Act. This law requires many agencies to provide a duplicate copy of the reports they submit to Congress to the Government Publishing Office, which maintains a central repository and makes the reports available to legislators and the public.

The legislation is a significant accomplishment for government transparency as the public would no longer need to file a FOIA request for these documents. It also is a significant accomplish for government accountability and efficiency, as all congressional offices would be able to obtain agency reports to Congress. In the past, the reports might easily be lost or not available to all.

Unfortunately, execution of the law faces significant technical challenges.

First, there is no canonical list of all reports due to Congress. The Clerk of the House of Representatives maintains a list, updated annually, of all reports due to Congress pursuant to House Rule II. Unfortunately, the Clerk interprets the cataloging requirement as applying only to reports the law requires be sent to the House or Senate, and not its committees or subcommittees. Accordingly, the catalog of mandated reports is incomplete, as many thousands of reports are directed to be sent to committees and thus do not show up on the Clerk’s list.

Many executive branch agencies maintain lists of the reports they believe they must provide to Congress. Unfortunately, there is no central repository of those reports, and the lists maintained by the agencies are not publicly available. As of FY 2026, agencies do provide a list of reports they believe are outdated or duplicative pursuant to the Eliminate Useless Reports Act of 2024, but publication of the duplicative and unnecessary reports in the agency’s Congressional Budget Justification does not solve the underlying problem.

Second, information about the reports are not available as structured data. The partial list, generated by the Clerk of the House (see example here), is published as a PDF. A file in PDF format is very difficult for a computer to read. In addition, the reports do not have a unique identifier — a social security number that only refers to a specific report. As a result, it would be challenging for GPO to look at the report they’ve received from an agency and to identify the same report from the list produced by the Clerk.

Third, there may be errors in the list produced by the Clerk or in the letter accompanying the document received by GPO. This makes it challenging to use citations to the law requiring the report be provided to Congress as a mechanism to determine a report has been provide.

The American Governance Institute has advocated for fixes to the ACMRA and Clerk’s report to address these issues. Fixes include expanding the Clerk’s report to include all reports due to Congress, requiring the Clerk’s report to be published as structured data and with unique identifiers for each report, and having GPO engage in quality assure to determine that they’ve received all the reports due to them.

Recently, however, we decided to try an additional approach: using technology to identify reports due to Congress and whether they have been received. To that end, we brought on government technology expert Dave Zvenyach to experiment with methods for sorting out the report collection and identifying what’s missing. Zvenyach presented preliminary findings of the project at the Congressional Data Task Force meeting in June, and he and Daniel had a more in-depth conversation about it recently that can be viewed below.

Among Dave’s findings:

• It is technologically possible to create a crosswalk between the reports identified in the Clerk’s report and those received by GPO.
• Of the nearly 3,300 mandated reports, GPO has received less than 1/3, or approximately 1,060 filings.
• GPO has received reports from agencies that are not in the Clerk’s list, approximately 400 reports so far.

Watch the video below, and watch his slides here.

The Library of Congress announced at the latest Congressional Data Task Force that the next public forum for Congress.gov will take place on September 24 from 1:00-3:00 PM. This forum is an opportunity for the public and the congressional community to provide feedback about the website.

Previous forums can be accessed on Congress.gov, and additional information will be available at the same link and on the Legislative Branch Innovation Hub.

The Congressional Data Task Force reconvened in a beautiful new location in the Jefferson Building at the Library of Congress on June 11. The meeting demonstrated how far CDTF has grown into a collaborative forum that facilitates legislative branch modernization. Most participating legislative branch support offices used their meeting slot to demo powerful platforms they’ve built for congressional users or preview coming attractions. We in civil society, working on the other side of the congressional firewall, were excited and appreciative that staff gave us a peek at their accomplishments through the demos, some of which were shared publicly for the first time.

There was one major data project unveiled at the meeting: the presentation by the Government Publishing Office has added the Legislative Branch Data Map compiled by the Congressional Data Coalition and the House Office of the Chief Administrative Officer in an easy-to-read format on the Legislative Branch Innovation Hub for all to explore. It is a catalog of datasets across the legislative branch that stakeholders can use to build their tools and solve their problems, something that heretofore has not existed. It grew out of a project demoed at the 2025 Congressional Hackathon and seeded by our 2023 “biased yet reliable guide to sources of information and data about Congress.”

The map includes the title, description, and link to the publisher source of data produced by legislative branch offices and civil society organizations as well as their DCAT, or standardized metadata. Like other GitHub pages, people can contribute new entries through this link. This resource centralizes dozens of data sources on a single reference and should be a great help to developers supporting legislative branch modernization. Veneice Smith in the Clerk’s office, who is managing the project, worked hard (in collaboration with others) to get the project to the point where the data is published online on the Innovation Hub.

We will have a longer write-up of information shared and tools demonstrated at the meeting, but here are a few highlights, including the text amendment project, Senate modernization, House modernization projects, and tracking congressionally mandated reports.

BILL TEXT ANALYSIS

Comparative Print Suite. New tools built on top of data standardization have made it possible for Hill staff to perform sophisticated bill text analysis with greater efficiency. The House Clerk’s Office demoed the completed full version of its comparative print suite platform, which CDTF coordinator Kirsten Gullickson noted is likely the first (but not only) artificial intelligence-enabled tool developed within the House. It cleanly shows inline text how a bill would change existing law and allows users to upload bill drafts. We’ve been tracking this complex, multi-year project and congratulate all involved. (We’ve written about this tool, also known as the Posey project, previously.)

It is available to all members of the House and is also used by CRS. A pilot group of about 80 Senate staff is using the comparative print suite as well, with interest growing from offices for more access. They have been working for a while to resolve licensing and administrative issues to allow for expanded usage in the Senate. (The comparative print suite we should note is not available to the public, although we have advocated for its availability.)

Text Analysis Program. Barbara Perkins from the Congressional Research Service provided a demo of its Text Analysis Program, which finds bills with similar text that may have been introduced in previous Congresses or have been tucked inside longer legislation. This is incredibly powerful, because it allows for the tracking of legislative ideas as they propagate across the current Congress or over multiple Congresses. This is version 2 of TAP, and it is very sophisticated indeed. (We had built a prototype of this approach, called BillMap, a few years back to demonstrate that it was possible to do.) TAP 1.0 was described by CRS in 2011 as a tool to identify similarities in congressional bills covering the 103rd Congress to date. Modernization efforts kicked off around FY 2022, according to the Library.

TAP 2.0 can help support resolving a recommendation by the House Modernization Committee (no. 120) to provide information resources that can show members’ activity on issues over time. In theory, it makes it possible for a member of Congress to receive credit for legislation they introduced even if that bill passes as part of a larger bill, or in an amended form. For that to work, the output of TAP must be shown on Congress.gov, identifying various kinds of relatedness of legislation.

Identifying related legislation is a tricky concept. How do you identify bills that are identical, near identical, and similar? Do you identify similarity in text, and if so, how similar do they need to be? Do you identify bills that have similar outcomes? We were comfortable with revealing how we identified similarity and having the user decide whether we made the right connection. CRS has some wariness about this approach, and so its analysis requires a lot more intervention by human experts who make the determination.

It is our view that combining the relatedness of legislation over multiple Congresses with information about those related bills, such as committee reports, votes, GAO reports, statements of administrative policy, and the like, could provide congressional users with a wealth of information to help them analyze legislation quickly by looking at how it has been responded to previously.

TAP 2.0 contains legislation going back to the 113th Congress and offers side-by-side views. TAP also is available in the House.

SENATE PROGRESS

Webster. Arin Shapiro of the Office of the Secretary of the Senate announced it will begin a revamp of the widely-used internal website called Webster. The technology behind the resource is becoming obsolete and the update will continue to help staff discover information useful in their work.1 An update involves several Senate stakeholders so it may be complex, but Shapiro said it should be completed in nine to 18 months. It is unclear whether it will be made available to House staff, but there is a growing effort to create greater federation of authentication between the two chambers to allow for better data and tool sharing.

The Bells. The secretary’s office also is finishing up development of a Senate version of an app that can keep users informed of open votes, quorum calls, and when the Senate is in session. For decades, this information has been displayed via lights on the clocks inside congressional office buildings and with alarms, transmitted through radio signals. It’s been a complicated technical challenge to convert this system to modern mobile devices, but the new app will work on and off the Hill via a VPN. The Senate team learned from a similar effort on the House side for its bells system. The data will be available internally through an API, but there are no plans as of yet to make it publicly available. There will be an app for internal users as well. Readers of this newsletter may remember Capitol Bells, an app developed by Ted Henderson that let the public know for whom the bells tolled, and had wide usage on the hill, too.

As we wrote about previously, the Secretary of the Senate’s office has updated the internal technology—the Amendment Tracking System—used to track senate bills and amendments as they’re considered on the Senate floor. As things stand, at times the Senate will vote on bills that have not yet been made publicly available.

Congress.gov data is far downstream from the Senate process, and unlike the House, which first publishes its legislation on docs.house.gov as uses it to measure online availability of legislation, there is no Senate equivalent. We have written letters and submitted testimony calling for legislation and amendments to be made publicly available as they are considered by the Senate. For its part, the Senate’s new technology is capable of providing public access, a testament to the robust design process for the new ATS, but leadership direction is required to provide transparency to the public.

MODERNIZATION FUND PROJECTS

CAO House Digital Service Director Ken Ward presented four projects that House appropriators recently approved to receive Modernization Initiative Account funding.

Constituent Engagement Innovation Project. This project will define a common format for information coming into member offices from constituent emails, phone calls, letters, etc. that go into offices’ constituent management systems software. The House Modernization Committee described the goal as “to provide Members with the option to select from a broad range of providers for the unique services that best meet their needs.” Starting this year, the HDS will engage in initial discovery and proof of concept work with the ultimate goal of producing a data lake so that members are not locked into a particular Constituent Management System, but can control their own data and move between platforms.

FlagTrack 2.0. Building on the successful flag tracking system, FlagTrack 2.0 will further streamline the process by allowing checks and refunds to be processed through the portal, providing an option for a flag to be sent to member offices or shipped directly to the constituent, and give constituents the ability to track their flag requests. This should take nine-18 months to complete.

Digital Signage and Wayfinding Initiative. Visiting the Capitol complex can be very confusing, and this project will improve the visitor experience in the House side of the complex. There will be nine new digital displays at entrances to the main House buildings that list member and public office locations as well as real time information about committee, floor, and public events. In addition, there will be a newly created House Visitors App that provides a 3D digital map, the ability to look up Member and public office locations, and turn-by-turn directions. You can scan a QR code and get directions. Think of it as Waze for Congress (well, the House side). The data to make all this work will be drawn from the CAO’s newish LegiDex directory, which provides up-to-date information about locations on the campus, as well as HouseCal, which keeps track of what events are occurring. This project should be complete in 18-24 months.

Service Academy Nomination Portal and Dashboard. Members of Congress annually nominate constituents to Service Academy positions, but the way member offices handle the process of collecting applications and processing recommendations varies widely. This tool will replace the spreadsheet and email-based systems used by most offices and replace it with a streamlined, online system. It will include an applicant portal, a public-facing web application for submitting applications and supporting documents and tracking application progress, and a staff dashboard, which allows staff to track and manage the applications. This is expected to be completed in nine-18 months.

Altogether, these projects should significantly improve the efficiency of office constituent services and the user experience for people interacting directly with Congress. This is a great validation of the process by which the Modernization Committee at House Administration recommends projects, appropriators provide the funds, and various stakeholders inside the House build the technology.

We are hopeful that once members can see their benefits, appropriators will restore the $10 million authorization for the Member Initiative Account so HDS can do even more. We also would be pleased to see more House-Senate and legislative branch-wide collaborations on modernizing technology, including seeing progress on implementing appropriators’ direction for the House and Senate to co-develop technology and sign a MOU to facilitate tool building. The House FY 2027 Legislative Branch Appropriations bill funds the account at $4 million for a second consecutive year.

ONGOING PROJECTS

There are a number of ongoing projects, including the joint House-Senate effort to provide unique IDs for lobbyists, the House’s building of a committee activity portal, and updates to USLM. More information about those projects will be published on the House Modernization Reports website in July.

MATCHING MANDATED REPORTS

Daniel shared a project the American Governance Institute has started with the help of government data and tech leader Dave Zvenyach to identify missing records of congressionally mandated reports that agencies are required to submit to GPO. It draws from a list of more than 3,000 required reports drawn up by the House Clerk’s Office and attempts to match it up with the more than 1,000 records on govinfo.gov. There is not a common data standard for what the legislative branch is ingesting and executive branch agencies are providing, which made the effort challenging.

Here are the key findings. First, most congressional mandates for reports to be provided to Congress do not have matching reports in the GPO repository. Second, 395 reports in the GPO repository are not identified in the House’s list of reports due to Congress. Finally, in light of metadata limitations arising from looking at the list of reports due against the list of reports received, the next step is to look at the transmission letters to Congress and match that against the reports themselves.

This is one of four technology projects undertaken by the American Governance Institute this year with the support of Zvenyach. Expect to see presentations on identifying absent House committee video data, transforming appropriations bills and reports into data, and tracking changes in appropriations report language.

At the meeting, Daniel also shared a few observations from a recent convening in Dublin of parliamentary officials from dozens of countries on data management and modernization efforts. He wrote a longer summary of the event on the Congressional Data Coalition blog.

The next CDTF meeting is scheduled for December 3, again in the Library of Congress. The next Congress.gov public forum will be on September 24. Several other events are in the works for September as well—this is when the Hackathon is usually scheduled—so stay tuned.

In addition, if you’re a staffer and interested in earning an ethics-approved professional certificate on ‘data skills for Congress,’ go here to learn more about the virtual class.

This conference touched on a wide variety of issues common to representative bodies, no matter their structure, relating to data management and the adoption of AI. More than sixty attendees from dozens of parliaments joined, including representatives from the United States. Participants included secretaries general — the senior civil servant in a chamber, roughly equivalent to the Clerk of the House — committee clerks, parliamentary counsels, legislative administrators, parliamentary technologists, and a handful of representatives from NGOs focused on parliamentary modernization.

The rules of the proceeding prevent me from sharing who said what. But I would like to share some of the ideas that I am taking away from the conference.

What Legislative Design Makes Possible

These international exchanges are valuable because they provide insight into the operations and decisions of our own legislatures. Some major challenges for Congress are settled policy or practice in other countries. Often, specific challenges like sustaining funding for modernization — or investments in the legislature generally — are a consequence of the design of the legislature itself. 

In some cases, administrative staff remain in office longer than the head of the legislature, while in others, that equivalent of our Speaker of the House is not a partisan position but an institutional-oriented one. Accordingly, staff may view themselves as internal advocates for the institution or merely as functionaries carrying out the wishes of political leaders. It may also matter a great deal whether the administrative staff is led by a single head, such as a secretary general, or by multiple officers.

Sustaining funding for parliamentary modernization also seems correlated with the political organization of the underlying representative democracy in whether a country has two major parties or more than two. 

Shared Projects, Shared Roadblocks

International events like this conference illustrate how there is a lot of overlap across the projects that parliaments try to build. The political and administrative arguments concerning adoption can be roadblocks others can learn from.

These off-the-record discussions also can surface the actual underlying issues hindering progress, which may be different from what is stated publicly.

For example, many parliaments use AI to transcribe their floor and committee proceedings. The arguments for doing so are speed, cost, transparency, and free accessibility of the information. Arguments made by those against AI-generated transcripts are often framed in terms of questionable accuracy. But accuracy is not always the real issue: Sometimes, members wish to retain the ability to alter their remarks before they are published. Other times, internal staff are afraid they will lose their jobs. Other countries’ experiences can point to ways to address both of those concerns and allow a transcription process to proceed.

Solving Parliamentary Problems in Parliamentary Ways

It is also interesting to see how parliaments solve their own unique problems. One parliament has a chamber in which its members vote by raising hands out of long-held tradition. It built a tool to recognize that practice and speed up voting. Another chamber had trouble with members voting on behalf of other members, or claiming presence for committee proceedings. To solve that problem, it uses a combination of fingerprint and facial recognition.

The AI Question

Unsurprisingly, how to deal with AI was a major topic of discussion in Dublin. Attendees all are struggling with when and how to bring AI into parliamentary processes, and with the reality that some members will use the technology beyond established guardrails or in the face of prohibitions.

The conference included optimistic and pessimistic views on the benefits of AI adoption by parliaments. Some AI optimists hoped the future will bring an AI-empowered legislature, in which the technology monitors official data sets and the news, identifies emerging problems, compiles research, and proposes solutions for a legislature to vote on. AI pessimists, however, worry about the concentration of power AI creates in society and its undercutting the legislature and short-circuiting the deliberative process. Others expressed skepticism about the reliability of AI products and concerns about them being economically and environmentally unsustainable.

Guardrails Are Not Enough

Although specific guardrails for AI use are important, reformers and modernizers also need to be thoughtful about the incorporation of AI into legislative work generally. Legislatures have been using forms of what we call artificial intelligence for decades, from spellcheck and OCR to natural language processing, however modern generative AI is a different animal.

Whether legislatures should build generative AI into their business processes risks creating dependencies on technology whose price may change dramatically. Think about what happened to Uber when the venture capital subsidies ended. It also raises questions of sovereignty — who owns the data, and is it kept secure — as well as questions of accuracy and externalities, including harms wrought on the environment. 

Should parliaments trust big technology companies to create tools that protect users and are in their – and therefore the public’s – best interests? Or should governments create publicly owned versions of these technologies?

Using AI to create deterministic tools, or parliamentary use of small language models instead of large ones, may help mitigate some of the major risks. For example, the comparative print suite, developed by the House of Representatives, uses natural language processing and its results are consistent over time. This kind of tool can be debugged and trusted to reach a reliable result. These are design questions facing legislatures, and they are very different from simply throwing everything up into Claude or ChatGPT.

The Value of Speaking Frankly Across Legislatures

The ability to get into granular detail about the challenges and opportunities around making our respective legislatures work — in a kind of convening that often works best in person and not for attribution — is a welcome opportunity for legislative senior staff and technology practitioners to keep up to speed with a rapidly changing environment.

The context of our era of change also is something parliamentary professionals hold in common. Our societies have seen a rise in tremendous economic inequality that allows the very wealthy to fund advocacy endlessly. We struggle with technology that supports the diffusion of information and misinformation with minimal human intervention, which challenges the very ability of legislatures to serve as deliberative policymaking bodies.

For this edition of Maker Space, we highlight a project launched by Candace Moix and Julie Lin, two alumnae of the TechCongress program, to provide a free, robust digital advocacy platform called yCapitol.ai. AGI supports leveling the informational playing field when it comes to Congress, so this is a very exciting project not only for under-resourced advocacy groups and nonprofits, but congressional staff, too.

Moix told me the idea came from watching Senate staff during her TechCongress fellowship interact with advocacy groups and industry lobbyists, noting the disparity not only of impact on legislative activity, but the understanding of how to connect successfully with the right congressional stakeholders (which is also why we co-developed this advocacy playbook for nonprofits). MyCapitol combines a variety of government data streams to allow users to track bills and issues as well as what topics outside groups, donors, and lobbyists are engaging on. It also integrates with Claude to create bill summaries, draft sample bill text, letters, and policy briefs, and summarize members’ career highlights and policy accomplishments.

Moix and Lin are upfront about how they are using generative AI not only to produce content but help build the platform. They have a thoughtful best practices guide on their website. It’s really about making the fundamentals of advocacy as efficient as possible for small organizations that may not otherwise have the time to engage with full rigor. The data analysis capacity, meanwhile, is intended to help identify the right champions and most promising pathways for legislative proposals, which professional lobbyists and senior staff know from years of experience but is a steep learning curve for everyone else.

As a civic tech project, MyCapitol actively invites user feedback. Moix is hopeful that congressional staff also will use it and discover new use cases to build out. There’s also a lot investigative journalists could do with the platform, too.

For this issue of Make Space, we highlight Capitol Labs’ comprehensive congressional hearing database. It’s the work of Jason Lemons, former VP at Prolegis and House staffer. Lemons described the project to me as an experiment in assembling a comprehensive set of data related to hearings under one hood. It builds upon what is available from the Library of Congress via Congress.gov APIs and committee videos to make the record of the hearing much more accessible and allow for analysis in ways that only otherwise exist on expensive subscription platforms.

The major innovation of the platform is in how it generates transcript data. Yes, Congress.gov offers official transcripts of hearings: but these take months (often many) to be processed. They also do not contain the metadata like timestamps and speaker tags that are needed for anything beyond Ctrl-F search. Capitol Labs uses AI to transcribe hearing video and construct the improved metadata, so users can see not only what was said, but track how long witnesses and members spoke, pull out relevant sections of text, and link to specific moments in the proceeding.

The beefier metadata also allows searching for keywords across committee hearings, written testimony, and CRS reports as well as track all the times particular members have spoken at hearings. It even displays all the hearings at which an individual witness has appeared before committees, something that congressional stakeholders are working on but is not yet available.

At the moment, the platform is limited by the processing time of the transcripts, but Lemons has a goal of increasing the pace to about 30 to 40 a week. This project is part of a larger development effort, but is currently free once a user creates an account.