Bringing AI to Capitol Hill: Why We Built Revere

On any given morning, an elected official’s inbox holds a veteran’s question about a delayed claim, a few hundred form emails from an advocacy campaign, and a note from a local scientist who knows more about the river running through the district than anyone in the building. All three types deserve an answer. Most offices want to give each of them answers, and most try; but the reality is, more often than not, the volume wins.

The Congressional Management Foundation (CMF) put numbers to the problem in 2022: in many offices, managing correspondence consumes about half of office resources, and form email advocacy campaigns account for 70 to 90 percent of the messages offices receive from constituents.

This problem has been brewing for decades. In 2004, the House and Senate combined received more than 200 million postal and email messages. House personal-office staffing, meanwhile, remained capped where a 1979 law set it, and Senate office sizes stayed roughly flat as well. In a 2010 CMF survey, when asked how long it would require for new text to be drafted, edited, and approved, 42 percent of offices estimated it took more than three weeks. That research is more than a decade old, but the imbalance it depicts has not gone away.

While mail itself is changing, offices are doing their best to keep up. In 2020, Cornell researchers sent more than 32,000 emails to roughly 7,000 state legislators; some written by people, while some were generated by AI. Offices answered both at nearly the same rate. The researchers framed the results as a warning: as AI writing tools spread, organized campaigns could arrive as thousands of individually worded messages, and offices may find it increasingly hard to tell orchestrated campaigns from individual voices.

Here at Civic, we’ve built Revere, an AI-powered all-in-one command center for government offices that handle constituent communications. Our goal is to bring modern, powerful technology to Capitol Hill and to every level of government. 

Where AI Fits In

Revere was built around AI from the beginning, in its data architecture as much as its features. Most of our AI-enabled features can be grouped into two categories: classification and generation. The classification features rely on machine learning models that we developed in-house, trained by subject matter experts on the team. The generative features support research, drafting, and retrieval of information. These call on a foundation model from within our secure account, and outputs are further honed by our internal training and supported by our data architecture and approach. 

The approach is grounded in what the industry calls retrieval-augmented generation: rather than generating text from memory, the system retrieves from a specific, verifiable body of knowledge and drafts from that. 

Think of it as the difference between an open-book and a closed-book exam. In a closed-book exam, you have to rely just on what you remember – which may lack nuance. In an open-book exam, you can reference the actual text directly, making answers more accurate and less likely to hallucinate. Retrieval-augmented generation is the open-book exam. For government offices, the ‘book’ is all your own records: past correspondence, approved positions and language, and casework history.

For the people running an office, the payoff is practical. Draft replies start from what the office has actually said before, not from a generic template. As an office’s record grows, staff can search that institutional memory in plain language. Drafts can be traced back to their sources before anything goes out. And no reply leaves the office without a person approving it.

From Processing Mail to Understanding It

Sorting is the easy part. The harder questions come after. What is the sentiment behind this week’s two thousand messages on a bill, and how does it vary across the district? When a constituent returns months after a casework file was closed with the same problem, can the caseworker find that history in the moment, or does it have to be rebuilt from scratch? How long does it take a staffer to research the background needed to answer the one-of-a-kind letter?

These are the questions we built Revere to answer. Using a machine learning model, messages are tagged and batched by topic as they arrive. The system preserves the context that raw counts leave behind, so a mail report can reflect the nuance in what constituents actually said rather than only how many said it. Casework and policy records stay connected to the individual, preserving continuity across time and staff turnover. AI-assisted research puts legislative background and policy summaries alongside the draft, shortening the distance between a question and an informed answer.

Built for Every Individual Who Writes, and Every Office That Answers

Offices receive messages from a range of people about a range of topics: Casework logs,  opinion letters, town hall follow-ups, stakeholder input, and the researcher or local official writing with expertise an office needs. Revere serves all of it, from Capitol Hill to a local city council, and it is nonpartisan by design. We provide the same tools and the same transparency for every office, regardless of party or issue.

Because this software carries communications between citizens and their government, security is foundational rather than optional. Civic has a strong security posture: we align our controls with NIST-800-53 guidelines, the framework federal agencies use to define system security; maintain an active SOC 2 Type II compliance, meaning an independent auditor has confirmed our security controls are operational over time; and receive annual penetration testing as well, demonstrating the robustness of our system. As part of those controls, data is encrypted at rest and in transit, hosted on secure U.S.-based cloud infrastructure. 

Looking Forward

The Congressional Management Foundation wrote in 2021 about rebuilding the democratic dialogue and argued that constituent engagement needed to be more responsive, more informed, and more trusted. We built Revere because we believe that the future is within reach, and that AI, applied carefully and reviewed by people, can help offices get there. When an office answers well, the person who wrote learns it was worth writing. That is the distance we are working to close.

If your office would like to see how it works, we would be glad to show you.